Leveraging Uncertainty in Machine Learning Accelerates Biological Discovery and Design

Neural network extrapolation to distant regions of the protein fitness landscape

Machine learning (ML) has transformed protein engineering by constructing models of the underlying sequence-function landscape to accelerate the discovery of new biomolecules. ML-guided protein design requires models, trained on local sequence-function information, to accurately predict distant fitness peaks. In this work, we evaluate neural networks' capacity to extrapolate beyond their training data. We perform model-guided design using a panel of neural network architectures trained on protein G (GB1)-Immunoglobulin G (IgG) binding data and experimentally test thousands of GB1 designs to systematically evaluate the models' extrapolation. We find each model architecture infers markedly different landscapes from the same data, which give rise to unique design preferences. We find simpler models excel in local extrapolation to design high fitness proteins, while more sophisticated convolutional models can venture deep into sequence space to design proteins that fold but are no longer functional. We also find that implementing a simple ensemble of convolutional neural networks enables robust design of high-performing variants in the local landscape. Our findings highlight how each architecture's inductive biases prime them to learn different aspects of the protein fitness landscape and how a simple ensembling approach makes protein engineering more robust.

Machine learning-guided co-optimization of fitness and diversity facilitates combinatorial library design in enzyme engineering

The effective design of combinatorial libraries to balance fitness and diversity facilitates the engineering of useful enzyme functions, particularly those that are poorly characterized or unknown in biology. We introduce MODIFY, a machine learning (ML) algorithm that learns from natural protein sequences to infer evolutionarily plausible mutations and predict enzyme fitness. MODIFY co-optimizes predicted fitness and sequence diversity of starting libraries, prioritizing high-fitness variants while ensuring broad sequence coverage. In silico evaluation shows that MODIFY outperforms state-of-the-art unsupervised methods in zero-shot fitness prediction and enables ML-guided directed evolution with enhanced efficiency. Using MODIFY, we engineer generalist biocatalysts derived from a thermostable cytochrome c to achieve enantioselective C-B and C-Si bond formation via a new-to-nature carbene transfer mechanism, leading to biocatalysts six mutations away from previously developed enzymes while exhibiting superior or comparable activities. These results demonstrate MODIFY's potential in solving challenging enzyme engineering problems beyond the reach of classic directed evolution.

Strain tracking with uncertainty quantification

The ability to detect and quantify microbiota over time has a plethora of clinical, basic science, and public health applications. One of the primary means of tracking microbiota is through sequencing technologies. When the microorganism of interest is well characterized or known a priori , targeted sequencing is often used. In many applications, however, untargeted bulk (shotgun) sequencing is more appropriate; for instance, the tracking of infection transmission events and nucleotide variants across multiple genomic loci, or studying the role of multiple genes in a particular phenotype. Given these applications, and the observation that pathogens (e.g. Clostridioides difficile, Escherichia coli, Salmonella enterica ) and other taxa of interest can reside at low relative abundance in the gastrointestinal tract, there is a critical need for algorithms that accurately track low-abundance taxa with strain level resolution. Here we present a sequence quality- and time-aware model, ChronoStrain , that introduces uncertainty quantification to gauge low-abundance species and significantly outperforms the current state-of-the-art on both real and synthetic data. ChronoStrain leverages sequences' quality scores and the samples' temporal information to produce a probability distribution over abundance trajectories for each strain tracked in the model. We demonstrate Chronostrain's improved performance in capturing post-antibiotic Escherichia coli strain blooms among women with recurrent urinary tract infections (UTIs) from the UTI Microbiome (UMB) Project. Other strain tracking models on the same data either show inconsistent temporal colonization or can only track consistently using very coarse groupings. In contrast, our probabilistic outputs can reveal the relationship between low-confidence strains present in the sample that cannot be reliably assigned a single reference label (either due to poor coverage or novelty) while simultaneously calling high-confidence strains that can be unambiguously assigned a label. We also analyze samples from the Early Life Microbiota Colonisation (ELMC) Study demonstrating the algorithm's ability to correctly identify Enterococcus faecalis strains using paired sample isolates as validation.

Reducing overconfident errors in molecular property classification using Posterior Network

Deep-learning-based classification models are increasingly used for predicting molecular properties in drug development. However, traditional classification models using the Softmax function often give overconfident mispredictions for out-of-distribution samples, highlighting a critical lack of accurate uncertainty estimation. Such limitations can result in substantial costs and should be avoided during drug development. Inspired by advances in evidential deep learning and Posterior Network, we replaced the Softmax function with a normalizing flow to enhance the uncertainty estimation ability of the model in molecular property classification. The proposed strategy was evaluated across diverse scenarios, including simulated experiments based on a synthetic dataset, ADMET predictions, and ligand-based virtual screening. The results demonstrate that compared with the vanilla model, the proposed strategy effectively alleviates the problem of giving overconfident but incorrect predictions. Our findings support the promising application of evidential deep learning in drug development and offer a valuable framework for further research.

Benchmarking compound activity prediction for real-world drug discovery applications

Identifying active compounds for target proteins is fundamental in early drug discovery. Recently, data-driven computational methods have demonstrated promising potential in predicting compound activities. However, there lacks a well-designed benchmark to comprehensively evaluate these methods from a practical perspective. To fill this gap, we propose a Compound Activity benchmark for Real-world Applications (CARA). Through carefully distinguishing assay types, designing train-test splitting schemes and selecting evaluation metrics, CARA can consider the biased distribution of current real-world compound activity data and avoid overestimation of model performances. We observed that although current models can make successful predictions for certain proportions of assays, their performances varied across different assays. In addition, evaluation of several few-shot training strategies demonstrated different performances related to task types. Overall, we provide a high-quality dataset for developing and evaluating compound activity prediction models, and the analyses in this work may inspire better applications of data-driven models in drug discovery.

A systematic analysis of regression models for protein engineering

To optimize proteins for particular traits holds great promise for industrial and pharmaceutical purposes. Machine Learning is increasingly applied in this field to predict properties of proteins, thereby guiding the experimental optimization process. A natural question is: How much progress are we making with such predictions, and how important is the choice of regressor and representation? In this paper, we demonstrate that different assessment criteria for regressor performance can lead to dramatically different conclusions, depending on the choice of metric, and how one defines generalization. We highlight the fundamental issues of sample bias in typical regression scenarios and how this can lead to misleading conclusions about regressor performance. Finally, we make the case for the importance of calibrated uncertainty in this domain.

Leveraging conformal prediction to annotate enzyme function space with limited false positives

Machine learning (ML) is increasingly being used to guide biological discovery in biomedicine such as prioritizing promising small molecules in drug discovery. In those applications, ML models are used to predict the properties of biological systems, and researchers use these predictions to prioritize candidates as new biological hypotheses for downstream experimental validations. However, when applied to unseen situations, these models can be overconfident and produce a large number of false positives. One solution to address this issue is to quantify the model's prediction uncertainty and provide a set of hypotheses with a controlled false discovery rate (FDR) pre-specified by researchers. We propose CPEC, an ML framework for FDR-controlled biological discovery. We demonstrate its effectiveness using enzyme function annotation as a case study, simulating the discovery process of identifying the functions of less-characterized enzymes. CPEC integrates a deep learning model with a statistical tool known as conformal prediction, providing accurate and FDR-controlled function predictions for a given protein enzyme. Conformal prediction provides rigorous statistical guarantees to the predictive model and ensures that the expected FDR will not exceed a user-specified level with high probability. Evaluation experiments show that CPEC achieves reliable FDR control, better or comparable prediction performance at a lower FDR than existing methods, and accurate predictions for enzymes under-represented in the training data. We expect CPEC to be a useful tool for biological discovery applications where a high yield rate in validation experiments is desired but the experimental budget is limited.

Discovery of potent inhibitors of α-synuclein aggregation using structure-based iterative learning

Machine learning methods hold the promise to reduce the costs and the failure rates of conventional drug discovery pipelines. This issue is especially pressing for neurodegenerative diseases, where the development of disease-modifying drugs has been particularly challenging. To address this problem, we describe here a machine learning approach to identify small molecule inhibitors of α-synuclein aggregation, a process implicated in Parkinson's disease and other synucleinopathies. Because the proliferation of α-synuclein aggregates takes place through autocatalytic secondary nucleation, we aim to identify compounds that bind the catalytic sites on the surface of the aggregates. To achieve this goal, we use structure-based machine learning in an iterative manner to first identify and then progressively optimize secondary nucleation inhibitors. Our results demonstrate that this approach leads to the facile identification of compounds two orders of magnitude more potent than previously reported ones.

BeatProfiler: Multimodal In Vitro Analysis of Cardiac Function Enables Machine Learning Classification of Diseases and Drugs

Goal: Contractile response and calcium handling are central to understanding cardiac function and physiology, yet existing methods of analysis to quantify these metrics are often time-consuming, prone to mistakes, or require specialized equipment/license. We developed BeatProfiler, a suite of cardiac analysis tools designed to quantify contractile function, calcium handling, and force generation for multiple in vitro cardiac models and apply downstream machine learning methods for deep phenotyping and classification. Methods: We first validate BeatProfiler's accuracy, robustness, and speed by benchmarking against existing tools with a fixed dataset. We further confirm its ability to robustly characterize disease and dose-dependent drug response. We then demonstrate that the data acquired by our automatic acquisition pipeline can be further harnessed for machine learning (ML) analysis to phenotype a disease model of restrictive cardiomyopathy and profile cardioactive drug functional response. To accurately classify between these biological signals, we apply feature-based ML and deep learning models (temporal convolutional-bidirectional long short-term memory model or TCN-BiLSTM). Results: Benchmarking against existing tools revealed that BeatProfiler detected and analyzed contraction and calcium signals better than existing tools through improved sensitivity in low signal data, reduction in false positives, and analysis speed increase by 7 to 50-fold. Of signals accurately detected by published methods (PMs), BeatProfiler's extracted features showed high correlations to PMs, confirming that it is reliable and consistent with PMs. The features extracted by BeatProfiler classified restrictive cardiomyopathy cardiomyocytes from isogenic healthy controls with 98% accuracy and identified relax90 as a top distinguishing feature in congruence with previous findings. We also show that our TCN-BiLSTM model was able to classify drug-free control and 4 cardiac drugs with different mechanisms of action at 96% accuracy. We further apply Grad-CAM on our convolution-based models to identify signature regions of perturbations by these drugs in calcium signals. Conclusions: We anticipate that the capabilities of BeatProfiler will help advance in vitro studies in cardiac biology through rapid phenotyping, revealing mechanisms underlying cardiac health and disease, and enabling objective classification of cardiac disease and responses to drugs.

Effective engineering of a ketoreductase for the biocatalytic synthesis of an ipatasertib precursor

Semi-rational enzyme engineering is a powerful method to develop industrial biocatalysts. Profiting from advances in molecular biology and bioinformatics, semi-rational approaches can effectively accelerate enzyme engineering campaigns. Here, we present the optimization of a ketoreductase from Sporidiobolus salmonicolor for the chemo-enzymatic synthesis of ipatasertib, a potent protein kinase B inhibitor. Harnessing the power of mutational scanning and structure-guided rational design, we created a 10-amino acid substituted variant exhibiting a 64-fold higher apparent kcat and improved robustness under process conditions compared to the wild-type enzyme. In addition, the benefit of algorithm-aided enzyme engineering was studied to derive correlations in protein sequence-function data, and it was found that the applied Gaussian processes allowed us to reduce enzyme library size. The final scalable and high performing biocatalytic process yielded the alcohol intermediate with ≥ 98% conversion and a diastereomeric excess of 99.7% (R,R-trans) from 100 g L-1 ketone after 30 h. Modelling and kinetic studies shed light on the mechanistic factors governing the improved reaction outcome, with mutations T134V, A238K, M242W and Q245S exerting the most beneficial effect on reduction activity towards the target ketone.

Opportunities and Challenges for Machine Learning-Assisted Enzyme Engineering

Enzymes can be engineered at the level of their amino acid sequences to optimize key properties such as expression, stability, substrate range, and catalytic efficiency-or even to unlock new catalytic activities not found in nature. Because the search space of possible proteins is vast, enzyme engineering usually involves discovering an enzyme starting point that has some level of the desired activity followed by directed evolution to improve its "fitness" for a desired application. Recently, machine learning (ML) has emerged as a powerful tool to complement this empirical process. ML models can contribute to (1) starting point discovery by functional annotation of known protein sequences or generating novel protein sequences with desired functions and (2) navigating protein fitness landscapes for fitness optimization by learning mappings between protein sequences and their associated fitness values. In this Outlook, we explain how ML complements enzyme engineering and discuss its future potential to unlock improved engineering outcomes.

Using Generative Modeling to Endow with Potency Initially Inert Compounds with Good Bioavailability and Low Toxicity

In the early stages of drug development, large chemical libraries are typically screened to identify compounds of promising potency against the chosen targets. Often, however, the resulting hit compounds tend to have poor drug metabolism and pharmacokinetics (DMPK), with negative developability features that may be difficult to eliminate. Therefore, starting the drug discovery process with a "null library", compounds that have highly desirable DMPK properties but no potency against the chosen targets, could be advantageous. Here, we explore the opportunities offered by machine learning to realize this strategy in the case of the inhibition of α-synuclein aggregation, a process associated with Parkinson's disease. We apply MolDQN, a generative machine learning method, to build an inhibitory activity against α-synuclein aggregation into an initial inactive compound with good DMPK properties. Our results illustrate how generative modeling can be used to endow initially inert compounds with desirable developability properties.

Efficient evolution of human antibodies from general protein language models

Natural evolution must explore a vast landscape of possible sequences for desirable yet rare mutations, suggesting that learning from natural evolutionary strategies could guide artificial evolution. Here we report that general protein language models can efficiently evolve human antibodies by suggesting mutations that are evolutionarily plausible, despite providing the model with no information about the target antigen, binding specificity or protein structure. We performed language-model-guided affinity maturation of seven antibodies, screening 20 or fewer variants of each antibody across only two rounds of laboratory evolution, and improved the binding affinities of four clinically relevant, highly mature antibodies up to sevenfold and three unmatured antibodies up to 160-fold, with many designs also demonstrating favorable thermostability and viral neutralization activity against Ebola and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pseudoviruses. The same models that improve antibody binding also guide efficient evolution across diverse protein families and selection pressures, including antibiotic resistance and enzyme activity, suggesting that these results generalize to many settings.

Meta learning addresses noisy and under-labeled data in machine learning-guided antibody engineering

Machine learning-guided protein engineering is rapidly progressing; however, collecting high-quality, large datasets remains a bottleneck. Directed evolution and protein engineering studies often require extensive experimental processes to eliminate noise and label protein sequence-function data. Meta learning has proven effective in other fields in learning from noisy data via bi-level optimization given the availability of a small dataset with trusted labels. Here, we leverage meta learning approaches to overcome noisy and under-labeled data and expedite workflows in antibody engineering. We generate yeast display antibody mutagenesis libraries and screen them for target antigen binding followed by deep sequencing. We then create representative learning tasks, including learning from noisy training data, positive and unlabeled learning, and learning out of distribution properties. We demonstrate that meta learning has the potential to reduce experimental screening time and improve the robustness of machine learning models by training with noisy and under-labeled training data.

Characterizing uncertainty in predictions of genomic sequence-to-activity models

Genomic sequence-to-activity models are increasingly utilized to understand gene regulatory syntax and probe the functional consequences of regulatory variation. Current models make accurate predictions of relative activity levels across the human reference genome, but their performance is more limited for predicting the effects of genetic variants, such as explaining gene expression variation across individuals. To better understand the causes of these shortcomings, we examine the uncertainty in predictions of genomic sequence-to-activity models using an ensemble of Basenji2 model replicates. We characterize prediction consistency on four types of sequences: reference genome sequences, reference genome sequences perturbed with TF motifs, eQTLs, and personal genome sequences. We observe that models tend to make high-confidence predictions on reference sequences, even when incorrect, and low-confidence predictions on sequences with variants. For eQTLs and personal genome sequences, we find that model replicates make inconsistent predictions in >50% of cases. Our findings suggest strategies to improve performance of these models.

Calibrated geometric deep learning improves kinase-drug binding predictions

Protein kinases regulate various cellular functions and hold significant pharmacological promise in cancer and other diseases. Although kinase inhibitors are one of the largest groups of approved drugs, much of the human kinome remains unexplored but potentially druggable. Computational approaches, such as machine learning, offer efficient solutions for exploring kinase-compound interactions and uncovering novel binding activities. Despite the increasing availability of three-dimensional (3D) protein and compound structures, existing methods predominantly focus on exploiting local features from one-dimensional protein sequences and two-dimensional molecular graphs to predict binding affinities, overlooking the 3D nature of the binding process. Here we present KDBNet, a deep learning algorithm that incorporates 3D protein and molecule structure data to predict binding affinities. KDBNet uses graph neural networks to learn structure representations of protein binding pockets and drug molecules, capturing the geometric and spatial characteristics of binding activity. In addition, we introduce an algorithm to quantify and calibrate the uncertainties of KDBNet's predictions, enhancing its utility in model-guided discovery in chemical or protein space. Experiments demonstrated that KDBNet outperforms existing deep learning models in predicting kinase-drug binding affinities. The uncertainties estimated by KDBNet are informative and well-calibrated with respect to prediction errors. When integrated with a Bayesian optimization framework, KDBNet enables data-efficient active learning and accelerates the exploration and exploitation of diverse high-binding kinase-drug pairs.

Neural network extrapolation to distant regions of the protein fitness landscape

Machine learning (ML) has transformed protein engineering by constructing models of the underlying sequence-function landscape to accelerate the discovery of new biomolecules. ML-guided protein design requires models, trained on local sequence-function information, to accurately predict distant fitness peaks. In this work, we evaluate neural networks' capacity to extrapolate beyond their training data. We perform model-guided design using a panel of neural network architectures trained on protein G (GB1)-Immunoglobulin G (IgG) binding data and experimentally test thousands of GB1 designs to systematically evaluate the models' extrapolation. We find each model architecture infers markedly different landscapes from the same data, which give rise to unique design preferences. We find simpler models excel in local extrapolation to design high fitness proteins, while more sophisticated convolutional models can venture deep into sequence space to design proteins that fold but are no longer functional. Our findings highlight how each architecture's inductive biases prime them to learn different aspects of the protein fitness landscape.

Machine Learning-Guided Protein Engineering

Recent progress in engineering highly promising biocatalysts has increasingly involved machine learning methods. These methods leverage existing experimental and simulation data to aid in the discovery and annotation of promising enzymes, as well as in suggesting beneficial mutations for improving known targets. The field of machine learning for protein engineering is gathering steam, driven by recent success stories and notable progress in other areas. It already encompasses ambitious tasks such as understanding and predicting protein structure and function, catalytic efficiency, enantioselectivity, protein dynamics, stability, solubility, aggregation, and more. Nonetheless, the field is still evolving, with many challenges to overcome and questions to address. In this Perspective, we provide an overview of ongoing trends in this domain, highlight recent case studies, and examine the current limitations of machine learning-based methods. We emphasize the crucial importance of thorough experimental validation of emerging models before their use for rational protein design. We present our opinions on the fundamental problems and outline the potential directions for future research.

RECOVER identifies synergistic drug combinations in vitro through sequential model optimization

For large libraries of small molecules, exhaustive combinatorial chemical screens become infeasible to perform when considering a range of disease models, assay conditions, and dose ranges. Deep learning models have achieved state-of-the-art results in silico for the prediction of synergy scores. However, databases of drug combinations are biased toward synergistic agents and results do not generalize out of distribution. During 5 rounds of experimentation, we employ sequential model optimization with a deep learning model to select drug combinations increasingly enriched for synergism and active against a cancer cell line-evaluating only ∼5% of the total search space. Moreover, we find that learned drug embeddings (using structural information) begin to reflect biological mechanisms. In silico benchmarking suggests search queries are ∼5-10× enriched for highly synergistic drug combinations by using sequential rounds of evaluation when compared with random selection or ∼3× when using a pretrained model.

Overfit deep neural network for predicting drug-target interactions

Drug-target interactions (DTIs) prediction is an important step in drug discovery. As traditional biological experiments or high-throughput screening are high cost and time-consuming, many deep learning models have been developed. Overfitting must be avoided when training deep learning models. We propose a simple framework, called OverfitDTI, for DTI prediction. In OverfitDTI, a deep neural network (DNN) model is overfit to sufficiently learn the features of the chemical space of drugs and the biological space of targets. The weights of trained DNN model form an implicit representation of the nonlinear relationship between drugs and targets. Performance of OverfitDTI on three public datasets showed that the overfit DNN models fit the nonlinear relationship with high accuracy. We identified fifteen compounds that interacted with TEK, a receptor tyrosine kinase contributing to vascular homeostasis, and the predicted AT9283 and dorsomorphin were experimentally demonstrated as inhibitors of TEK in human umbilical vein endothelial cells (HUVECs).

Efforts to Minimise the Bacterial Genome as a Free-Living Growing System

Exploring the minimal genetic requirements for cells to maintain free living is an exciting topic in biology. Multiple approaches are employed to address the question of the minimal genome. In addition to constructing the synthetic genome in the test tube, reducing the size of the wild-type genome is a practical approach for obtaining the essential genomic sequence for living cells. The well-studied Escherichia coli has been used as a model organism for genome reduction owing to its fast growth and easy manipulation. Extensive studies have reported how to reduce the bacterial genome and the collections of genomic disturbed strains acquired, which were sufficiently reviewed previously. However, the common issue of growth decrease caused by genetic disturbance remains largely unaddressed. This mini-review discusses the considerable efforts made to improve growth fitness, which was decreased due to genome reduction. The proposal and perspective are clarified for further accumulated genetic deletion to minimise the Escherichia coli genome in terms of genome reduction, experimental evolution, medium optimization, and machine learning.

Active learning of enhancer and silencer regulatory grammar in photoreceptors

Cis-regulatory elements (CREs) direct gene expression in health and disease, and models that can accurately predict their activities from DNA sequences are crucial for biomedicine. Deep learning represents one emerging strategy to model the regulatory grammar that relates CRE sequence to function. However, these models require training data on a scale that exceeds the number of CREs in the genome. We address this problem using active machine learning to iteratively train models on multiple rounds of synthetic DNA sequences assayed in live mammalian retinas. During each round of training the model actively selects sequence perturbations to assay, thereby efficiently generating informative training data. We iteratively trained a model that predicts the activities of sequences containing binding motifs for the photoreceptor transcription factor Cone-rod homeobox (CRX) using an order of magnitude less training data than current approaches. The model's internal confidence estimates of its predictions are reliable guides for designing sequences with high activity. The model correctly identified critical sequence differences between active and inactive sequences with nearly identical transcription factor binding sites, and revealed order and spacing preferences for combinations of motifs. Our results establish active learning as an effective method to train accurate deep learning models of cis-regulatory function after exhausting naturally occurring training examples in the genome.

In vitro continuous protein evolution empowered by machine learning and automation

Directed evolution has become one of the most successful and powerful tools for protein engineering. However, the efforts required for designing, constructing, and screening a large library of variants can be laborious, time-consuming, and costly. With the recent advent of machine learning (ML) in the directed evolution of proteins, researchers can now evaluate variants in silico and guide a more efficient directed evolution campaign. Furthermore, recent advancements in laboratory automation have enabled the rapid execution of long, complex experiments for high-throughput data acquisition in both industrial and academic settings, thus providing the means to collect a large quantity of data required to develop ML models for protein engineering. In this perspective, we propose a closed-loop in vitro continuous protein evolution framework that leverages the best of both worlds, ML and automation, and provide a brief overview of the recent developments in the field.

Linear-Scaling Kernels for Protein Sequences and Small Molecules Outperform Deep Learning While Providing Uncertainty Quantitation and Improved Interpretability

Gaussian process (GP) is a Bayesian model which provides several advantages for regression tasks in machine learning such as reliable quantitation of uncertainty and improved interpretability. Their adoption has been precluded by their excessive computational cost and by the difficulty in adapting them for analyzing sequences (e.g., amino acid sequences) and graphs (e.g., small molecules). In this study, we introduce a group of random feature-approximated kernels for sequences and graphs that exhibit linear scaling with both the size of the training set and the size of the sequences or graphs. We incorporate these new kernels into our new Python library for GP regression, xGPR, and develop an efficient and scalable algorithm for fitting GPs equipped with these kernels to large datasets. We compare the performance of xGPR on 17 different benchmarks with both standard and state-of-the-art deep learning models and find that GP regression achieves highly competitive accuracy for these tasks while providing with well-calibrated uncertainty quantitation and improved interpretability. Finally, in a simple experiment, we illustrate how xGPR may be used as part of an active learning strategy to engineer a protein with a desired property in an automated way without human intervention.

Understanding the host-pathogen evolutionary balance through Gaussian process modeling of SARS-CoV-2

We have developed a machine learning (ML) approach using Gaussian process (GP)-based spatial covariance (SCV) to track the impact of spatial-temporal mutational events driving host-pathogen balance in biology. We show how SCV can be applied to understanding the response of evolving covariant relationships linking the variant pattern of virus spread to pathology for the entire SARS-CoV-2 genome on a daily basis. We show that GP-based SCV relationships in conjunction with genome-wide co-occurrence analysis provides an early warning anomaly detection (EWAD) system for the emergence of variants of concern (VOCs). EWAD can anticipate changes in the pattern of performance of spread and pathology weeks in advance, identifying signatures destined to become VOCs. GP-based analyses of variation across entire viral genomes can be used to monitor micro and macro features responsible for host-pathogen balance. The versatility of GP-based SCV defines starting point for understanding nature's evolutionary path to complexity through natural selection.

Contrastive learning in protein language space predicts interactions between drugs and protein targets

Sequence-based prediction of drug-target interactions has the potential to accelerate drug discovery by complementing experimental screens. Such computational prediction needs to be generalizable and scalable while remaining sensitive to subtle variations in the inputs. However, current computational techniques fail to simultaneously meet these goals, often sacrificing performance of one to achieve the others. We develop a deep learning model, ConPLex, successfully leveraging the advances in pretrained protein language models ("PLex") and employing a protein-anchored contrastive coembedding ("Con") to outperform state-of-the-art approaches. ConPLex achieves high accuracy, broad adaptivity to unseen data, and specificity against decoy compounds. It makes predictions of binding based on the distance between learned representations, enabling predictions at the scale of massive compound libraries and the human proteome. Experimental testing of 19 kinase-drug interaction predictions validated 12 interactions, including four with subnanomolar affinity, plus a strongly binding EPHB1 inhibitor (KD = 1.3 nM). Furthermore, ConPLex embeddings are interpretable, which enables us to visualize the drug-target embedding space and use embeddings to characterize the function of human cell-surface proteins. We anticipate that ConPLex will facilitate efficient drug discovery by making highly sensitive in silico drug screening feasible at the genome scale. ConPLex is available open source at https://ConPLex.csail.mit.edu.

Next-Generation Adjuvants: Applying Engineering Methods to Create and Evaluate Novel Immunological Responses

Adjuvants are a critical component of vaccines. Adjuvants typically target receptors that activate innate immune signaling pathways. Historically, adjuvant development has been laborious and slow, but has begun to accelerate over the past decade. Current adjuvant development consists of screening for an activating molecule, formulating lead molecules with an antigen, and testing this combination in an animal model. There are very few adjuvants approved for use in vaccines, however, as new candidates often fail due to poor clinical efficacy, intolerable side effects, or formulation limitations. Here, we consider new approaches using tools from engineering to improve next-generation adjuvant discovery and development. These approaches will create new immunological outcomes that will be evaluated with novel diagnostic tools. Potential improved immunological outcomes include reduced vaccine reactogenicity, tunable adaptive responses, and enhanced adjuvant delivery. Evaluations of these outcomes can leverage computational approaches to interpret "big data" obtained from experimentation. Applying engineering concepts and solutions will provide alternative perspectives, further accelerating the field of adjuvant discovery.

Evaluating point-prediction uncertainties in neural networks for protein-ligand binding prediction

Neural Network (NN) models provide potential to speed up the drug discovery process and reduce its failure rates. The success of NN models requires uncertainty quantification (UQ) as drug discovery explores chemical space beyond the training data distribution. Standard NN models do not provide uncertainty information. Some methods require changing the NN architecture or training procedure, limiting the selection of NN models. Moreover, predictive uncertainty can come from different sources. It is important to have the ability to separately model different types of predictive uncertainty, as the model can take assorted actions depending on the source of uncertainty. In this paper, we examine UQ methods that estimate different sources of predictive uncertainty for NN models aiming at protein-ligand binding prediction. We use our prior knowledge on chemical compounds to design the experiments. By utilizing a visualization method we create non-overlapping and chemically diverse partitions from a collection of chemical compounds. These partitions are used as training and test set splits to explore NN model uncertainty. We demonstrate how the uncertainties estimated by the selected methods describe different sources of uncertainty under different partitions and featurization schemes and the relationship to prediction error.

Artificial intelligence for compound pharmacokinetics prediction

Optimisation of compound pharmacokinetics (PK) is an integral part of drug discovery and development. Animal in vivo PK data as well as human and animal in vitro systems are routinely utilised to evaluate PK in humans. In recent years machine learning and artificial intelligence (AI) emerged as a major tool for modelling of in vivo animal and human PK, enabling prediction from chemical structure early in drug discovery, and therefore offering opportunities to guide the design and prioritisation of molecules based on relevant in vivo properties and, ultimately, predicting human PK at the point of design. This review presents recent advances in machine learning and AI models for in vivo animal and human PK for small-molecule compounds as well as some examples for antibody therapeutics.

Multilingual translation for zero-shot biomedical classification using BioTranslator

Existing annotation paradigms rely on controlled vocabularies, where each data instance is classified into one term from a predefined set of controlled vocabularies. This paradigm restricts the analysis to concepts that are known and well-characterized. Here, we present the novel multilingual translation method BioTranslator to address this problem. BioTranslator takes a user-written textual description of a new concept and then translates this description to a non-text biological data instance. The key idea of BioTranslator is to develop a multilingual translation framework, where multiple modalities of biological data are all translated to text. We demonstrate how BioTranslator enables the identification of novel cell types using only a textual description and how BioTranslator can be further generalized to protein function prediction and drug target identification. Our tool frees scientists from limiting their analyses within predefined controlled vocabularies, enabling them to interact with biological data using free text.

Advancing Targeted Protein Degradation via Multiomics Profiling and Artificial Intelligence

Only around 20% of the human proteome is considered to be druggable with small-molecule antagonists. This leaves some of the most compelling therapeutic targets outside the reach of ligand discovery. The concept of targeted protein degradation (TPD) promises to overcome some of these limitations. In brief, TPD is dependent on small molecules that induce the proximity between a protein of interest (POI) and an E3 ubiquitin ligase, causing ubiquitination and degradation of the POI. In this perspective, we want to reflect on current challenges in the field, and discuss how advances in multiomics profiling, artificial intelligence, and machine learning (AI/ML) will be vital in overcoming them. The presented roadmap is discussed in the context of small-molecule degraders but is equally applicable for other emerging proximity-inducing modalities.

Protein engineering via Bayesian optimization-guided evolutionary algorithm and robotic experiments

Directed protein evolution applies repeated rounds of genetic mutagenesis and phenotypic screening and is often limited by experimental throughput. Through in silico prioritization of mutant sequences, machine learning has been applied to reduce wet lab burden to a level practical for human researchers. On the other hand, robotics permits large batches and rapid iterations for protein engineering cycles, but such capacities have not been well exploited in existing machine learning-assisted directed evolution approaches. Here, we report a scalable and batched method, Bayesian Optimization-guided EVOlutionary (BO-EVO) algorithm, to guide multiple rounds of robotic experiments to explore protein fitness landscapes of combinatorial mutagenesis libraries. We first examined various design specifications based on an empirical landscape of protein G domain B1. Then, BO-EVO was successfully generalized to another empirical landscape of an Escherichia coli kinase PhoQ, as well as simulated NK landscapes with up to moderate epistasis. This approach was then applied to guide robotic library creation and screening to engineer enzyme specificity of RhlA, a key biosynthetic enzyme for rhamnolipid biosurfactants. A 4.8-fold improvement in producing a target rhamnolipid congener was achieved after examining less than 1% of all possible mutants after four iterations. Overall, BO-EVO proves to be an efficient and general approach to guide combinatorial protein engineering without prior knowledge.

Learning with uncertainty to accelerate the discovery of histone lysine-specific demethylase 1A (KDM1A/LSD1) inhibitors

Machine learning including modern deep learning models has been extensively used in drug design and screening. However, reliable prediction of molecular properties is still challenging when exploring out-of-domain regimes, even for deep neural networks. Therefore, it is important to understand the uncertainty of model predictions, especially when the predictions are used to guide further experiments. In this study, we explored the utility and effectiveness of evidential uncertainty in compound screening. The evidential Graphormer model was proposed for uncertainty-guided discovery of KDM1A/LSD1 inhibitors. The benchmarking results illustrated that (i) Graphormer exhibited comparative predictive power to state-of-the-art models, and (ii) evidential regression enabled well-ranked uncertainty estimates and calibrated predictions. Subsequently, we leveraged time-splitting on the curated KDM1A/LSD1 dataset to simulate out-of-distribution predictions. The retrospective virtual screening showed that the evidential uncertainties helped reduce false positives among the top-acquired compounds and thus enabled higher experimental validation rates. The trained model was then used to virtually screen an independent in-house compound set. The top 50 compounds ranked by two different ranking strategies were experimentally validated, respectively. In general, our study highlighted the importance to understand the uncertainty in prediction, which can be recognized as an interpretable dimension to model predictions.

Antibiotic discovery in the artificial intelligence era

As the global burden of antibiotic resistance continues to grow, creative approaches to antibiotic discovery are needed to accelerate the development of novel medicines. A rapidly progressing computational revolution-artificial intelligence-offers an optimistic path forward due to its ability to alleviate bottlenecks in the antibiotic discovery pipeline. In this review, we discuss how advancements in artificial intelligence are reinvigorating the adoption of past antibiotic discovery models-namely natural product exploration and small molecule screening. We then explore the application of contemporary machine learning approaches to emerging areas of antibiotic discovery, including antibacterial systems biology, drug combination development, antimicrobial peptide discovery, and mechanism of action prediction. Lastly, we propose a call to action for open access of high-quality screening datasets and interdisciplinary collaboration to accelerate the rate at which machine learning models can be trained and new antibiotic drugs can be developed.

Genome-wide mapping of somatic mutation rates uncovers drivers of cancer

Identification of cancer driver mutations that confer a proliferative advantage is central to understanding cancer; however, searches have often been limited to protein-coding sequences and specific non-coding elements (for example, promoters) because of the challenge of modeling the highly variable somatic mutation rates observed across tumor genomes. Here we present Dig, a method to search for driver elements and mutations anywhere in the genome. We use deep neural networks to map cancer-specific mutation rates genome-wide at kilobase-scale resolution. These estimates are then refined to search for evidence of driver mutations under positive selection throughout the genome by comparing observed to expected mutation counts. We mapped mutation rates for 37 cancer types and applied these maps to identify putative drivers within intronic cryptic splice regions, 5' untranslated regions and infrequently mutated genes. Our high-resolution mutation rate maps, available for web-based exploration, are a resource to enable driver discovery genome-wide.

Uncertainty quantification: Can we trust artificial intelligence in drug discovery?

The problem of human trust is one of the most fundamental problems in applied artificial intelligence in drug discovery. In silico models have been widely used to accelerate the process of drug discovery in recent years. However, most of these models can only give reliable predictions within a limited chemical space that the training set covers (applicability domain). Predictions of samples falling outside the applicability domain are unreliable and sometimes dangerous for the drug-design decision-making process. Uncertainty quantification accordingly has drawn great attention to enable autonomous drug designing. By quantifying the confidence level of model predictions, the reliability of the predictions can be quantitatively represented to assist researchers in their molecular reasoning and experimental design. Here we summarize the state-of-the-art approaches to uncertainty quantification and underline how they can be used for drug design and discovery projects. Furthermore, we also outline four representative application scenarios of uncertainty quantification in drug discovery.

Protease Activity Analysis: A Toolkit for Analyzing Enzyme Activity Data

Analyzing the activity of proteases and their substrates is critical to defining the biological functions of these enzymes and to designing new diagnostics and therapeutics that target protease dysregulation in disease. While a wide range of databases and algorithms have been created to better predict protease cleavage sites, there is a dearth of computational tools to automate analysis of in vitro and in vivo protease assays. This necessitates individual researchers to develop their own analytical pipelines, resulting in a lack of standardization across the field. To facilitate protease research, here we present Protease Activity Analysis (PAA), a toolkit for the preprocessing, visualization, machine learning analysis, and querying of protease activity data sets. PAA leverages a Python-based object-oriented implementation that provides a modular framework for streamlined analysis across three major components. First, PAA provides a facile framework to query data sets of synthetic peptide substrates and their cleavage susceptibilities across a diverse set of proteases. To complement the database functionality, PAA also includes tools for the automated analysis and visualization of user-input enzyme-substrate activity measurements generated through in vitro screens against synthetic peptide substrates. Finally, PAA supports a set of modular machine learning functions to analyze in vivo protease activity signatures that are generated by activity-based sensors. Overall, PAA offers the protease community a breadth of computational tools to streamline research, taking a step toward standardizing data analysis across the field and in chemical biology and biochemistry at large.

Deep generative models for peptide design

Computers can already be programmed for superhuman pattern recognition of images and text. For machines to discover novel molecules, they must first be trained to sort through the many characteristics of molecules and determine which properties should be retained, suppressed, or enhanced to optimize functions of interest. Machines need to be able to understand, read, write, and eventually create new molecules. Today, this creative process relies on deep generative models, which have gained popularity since powerful deep neural networks were introduced to generative model frameworks. In recent years, they have demonstrated excellent ability to model complex distribution of real-word data (e.g., images, audio, text, molecules, and biological sequences). Deep generative models can generate data beyond those provided in training samples, thus yielding an efficient and rapid tool for exploring the massive search space of high-dimensional data such as DNA/protein sequences and facilitating the design of biomolecules with desired functions. Here, we review the emerging field of deep generative models applied to peptide science. In particular, we discuss several popular deep generative model frameworks as well as their applications to generate peptides with various kinds of properties (e.g., antimicrobial, anticancer, cell penetration, etc). We conclude our review with a discussion of current limitations and future perspectives in this emerging field.

Triangulating variation in the population to define mechanisms for precision management of genetic disease

To understand mechanistically how the protein fold is shaped by therapeutics to inform precision management of disease, we developed variation-capture (VarC) mapping. VarC triangulates sparse sequence variation information found in the population using Gaussian process regression (GPR)-based machine learning to define the combined pairwise-residue interactions contributing to dynamic protein function in the individual in response to therapeutics. Using VarC mapping, we now reveal the pairwise-residue covariant relationships across the entire protein fold of cystic fibrosis (CF) transmembrane conductance regulator (CFTR) to define the molecular mechanisms of clinically approved CF chemical modulators. We discover an energetically destabilized covariant core containing a di-acidic YKDAD endoplasmic reticulum (ER) exit code that is only weakly corrected by current therapeutics. Our results illustrate that VarC provides a generalizable tool to triangulate information from genetic variation in the population to mechanistically discover therapeutic strategies that guide precision management of the individual.

Prediction of In Vivo Pharmacokinetic Parameters and Time-Exposure Curves in Rats Using Machine Learning from the Chemical Structure

Animal pharmacokinetic (PK) data as well as human and animal in vitro systems are utilized in drug discovery to define the rate and route of drug elimination. Accurate prediction and mechanistic understanding of drug clearance and disposition in animals provide a degree of confidence for extrapolation to humans. In addition, prediction of in vivo properties can be used to improve design during drug discovery, help select compounds with better properties, and reduce the number of in vivo experiments. In this study, we generated machine learning models able to predict rat in vivo PK parameters and concentration-time PK profiles based on the molecular chemical structure and either measured or predicted in vitro parameters. The models were trained on internal in vivo rat PK data for over 3000 diverse compounds from multiple projects and therapeutic areas, and the predicted endpoints include clearance and oral bioavailability. We compared the performance of various traditional machine learning algorithms and deep learning approaches, including graph convolutional neural networks. The best models for PK parameters achieved R² = 0.63 [root mean squared error (RMSE) = 0.26] for clearance and R² = 0.55 (RMSE = 0.46) for bioavailability. The models provide a fast and cost-efficient way to guide the design of molecules with optimal PK profiles, to enable the prediction of virtual compounds at the point of design, and to drive prioritization of compounds for in vivo assays.

The long road to calibrated prediction uncertainty in computational chemistry

Uncertainty quantification (UQ) in computational chemistry (CC) is still in its infancy. Very few CC methods are designed to provide a confidence level on their predictions, and most users still rely improperly on the mean absolute error as an accuracy metric. The development of reliable UQ methods is essential, notably for CC to be used confidently in industrial processes. A review of the CC-UQ literature shows that there is no common standard procedure to report or validate prediction uncertainty. I consider here analysis tools using concepts (calibration and sharpness) developed in meteorology and machine learning for the validation of probabilistic forecasters. These tools are adapted to CC-UQ and applied to datasets of prediction uncertainties provided by composite methods, Bayesian ensembles methods, and machine learning and a posteriori statistical methods.

Machine learning modeling of family wide enzyme-substrate specificity screens

Biocatalysis is a promising approach to sustainably synthesize pharmaceuticals, complex natural products, and commodity chemicals at scale. However, the adoption of biocatalysis is limited by our ability to select enzymes that will catalyze their natural chemical transformation on non-natural substrates. While machine learning and in silico directed evolution are well-posed for this predictive modeling challenge, efforts to date have primarily aimed to increase activity against a single known substrate, rather than to identify enzymes capable of acting on new substrates of interest. To address this need, we curate 6 different high-quality enzyme family screens from the literature that each measure multiple enzymes against multiple substrates. We compare machine learning-based compound-protein interaction (CPI) modeling approaches from the literature used for predicting drug-target interactions. Surprisingly, comparing these interaction-based models against collections of independent (single task) enzyme-only or substrate-only models reveals that current CPI approaches are incapable of learning interactions between compounds and proteins in the current family level data regime. We further validate this observation by demonstrating that our no-interaction baseline can outperform CPI-based models from the literature used to guide the discovery of kinase inhibitors. Given the high performance of non-interaction based models, we introduce a new structure-based strategy for pooling residue representations across a protein sequence. Altogether, this work motivates a principled path forward in order to build and evaluate meaningful predictive models for biocatalysis and other drug discovery applications.

Predicting interactions between compounds and proteins represents a long-standing dream of drug discovery and protein engineering. Robust models of enzyme-substrate scope would dramatically advance our ability to design synthetic routes involving enzymatic catalysis. However, the lack of standardization between compound-protein interaction studies makes it difficult to evaluate the generalizability of such models. In this work we take a critical step forward by standardizing high-quality datasets measuring enzyme-substrate interactions, outlining rigorous evaluations, and proposing a new way to integrate structural information into protein representations. In testing previous modeling approaches, we highlight a surprising inability of existing models to effectively leverage compound-protein interactions to improve generalization, which challenges a perception in the literature. This establishes future opportunities for model development and integration of enzyme-substrate scope models into computer-aided synthesis planning software.

Evaluating Pointwise Reliability of Machine Learning prediction

Interest in Machine Learning applications to tackle clinical and biological problems is increasing. This is driven by promising results reported in many research papers, the increasing number of AI-based software products, and by the general interest in Artificial Intelligence to solve complex problems. It is therefore of importance to improve the quality of machine learning output and add safeguards to support their adoption. In addition to regulatory and logistical strategies, a crucial aspect is to detect when a Machine Learning model is not able to generalize to new unseen instances, which may originate from a population distant to that of the training population or from an under-represented subpopulation. As a result, the prediction of the machine learning model for these instances may be often wrong, given that the model is applied outside its "reliable" space of work, leading to a decreasing trust of the final users, such as clinicians. For this reason, when a model is deployed in practice, it would be important to advise users when the model's predictions may be unreliable, especially in high-stakes applications, including those in healthcare. Yet, reliability assessment of each machine learning prediction is still poorly addressed. Here, we review approaches that can support the identification of unreliable predictions, we harmonize the notation and terminology of relevant concepts, and we highlight and extend possible interrelationships and overlap among concepts. We then demonstrate, on simulated and real data for ICU in-hospital death prediction, a possible integrative framework for the identification of reliable and unreliable predictions. To do so, our proposed approach implements two complementary principles, namely the density principle and the local fit principle. The density principle verifies that the instance we want to evaluate is similar to the training set. The local fit principle verifies that the trained model performs well on training subsets that are more similar to the instance under evaluation. Our work can contribute to consolidating work in machine learning especially in medicine.

Learning Strategies in Protein Directed Evolution

Synthetic biology is a fast-evolving research field that combines biology and engineering principles to develop new biological systems for medical, pharmacological, and industrial applications. Synthetic biologists use iterative "design, build, test, and learn" cycles to efficiently engineer genetic systems that are reliable, reproducible, and predictable. Protein engineering by directed evolution can benefit from such a systematic engineering approach for various reasons. Learning can be carried out before starting, throughout or after finalizing a directed evolution project. Computational tools, bioinformatics, and scanning mutagenesis methods can be excellent starting points, while molecular dynamics simulations and other strategies can guide engineering efforts. Similarly, studying protein intermediates along evolutionary pathways offers fascinating insights into the molecular mechanisms shaped by evolution. The learning step of the cycle is not only crucial for proteins or enzymes that are not suitable for high-throughput screening or selection systems, but it is also valuable for any platform that can generate a large amount of data that can be aided by machine learning algorithms. The main challenge in protein engineering is to predict the effect of a single mutation on one functional parameter-to say nothing of several mutations on multiple parameters. This is largely due to nonadditive mutational interactions, known as epistatic effects-beneficial mutations present in a genetic background may not be beneficial in another genetic background. In this work, we provide an overview of experimental and computational strategies that can guide the user to learn protein function at different stages in a directed evolution project. We also discuss how epistatic effects can influence the success of directed evolution projects. Since machine learning is gaining momentum in protein engineering and the field is becoming more interdisciplinary thanks to collaboration between mathematicians, computational scientists, engineers, molecular biologists, and chemists, we provide a general workflow that familiarizes nonexperts with the basic concepts, dataset requirements, learning approaches, model capabilities and performance metrics of this intriguing area. Finally, we also provide some practical recommendations on how machine learning can harness epistatic effects for engineering proteins in an "outside-the-box" way.

Machine learning & deep learning in data-driven decision making of drug discovery and challenges in high-quality data acquisition in the pharmaceutical industry

Predicting novel small molecule bioactivities for the target deconvolution, hit-to-lead optimization in drug discovery research, requires molecular representation. Previous reports have demonstrated that machine learning (ML) and deep learning (DL) have substantial implications in virtual screening, peptide synthesis, drug ADMET screening and biomarker discovery. These strategies can increase the positive outcomes in the drug discovery process without false-positive rates and can be achieved in a cost-effective way with a minimum duration of time by high-quality data acquisition. This review substantially discusses the recent updates in AI tools as cheminformatics application in medicinal chemistry for the data-driven decision making of drug discovery and challenges in high-quality data acquisition in the pharmaceutical industry while improving small-molecule bioactivities and properties.

Calibrated uncertainty for molecular property prediction using ensembles of message passing neural networks

Data-driven methods based on machine learning have the potential to accelerate computational analysis of atomic structures. In this context, reliable uncertainty estimates are important for assessing confidence in predictions and enabling decision making. However, machine learning models can produce badly calibrated uncertainty estimates and it is therefore crucial to detect and handle uncertainty carefully. In this work we extend a message passing neural network designed specifically for predicting properties of molecules and materials with a calibrated probabilistic predictive distribution. The method presented in this paper differs from previous work by considering both aleatoric and epistemic uncertainty in a unified framework, and by recalibrating the predictive distribution on unseen data. Through computer experiments, we show that our approach results in accurate models for predicting molecular formation energies with well calibrated uncertainty in and out of the training data distribution on two public molecular benchmark datasets, QM9 and PC9. The proposed method provides a general framework for training and evaluating neural network ensemble models that are able to produce accurate predictions of properties of molecules with well calibrated uncertainty estimates.

Impact of Artificial Intelligence on Compound Discovery, Design, and Synthesis

As in other areas, artificial intelligence (AI) is heavily promoted in different scientific fields, including chemistry. Although chemistry traditionally tends to be a conservative field and slower than others to adapt new concepts, AI is increasingly being investigated across chemical disciplines. In medicinal chemistry, supported by computer-aided drug design and cheminformatics, computational methods have long been employed to aid in the search for and optimization of active compounds. We are currently witnessing a multitude of AI-related publications in the medicinal-chemistry-relevant literature and anticipate that the numbers will further increase. Often, advances through AI promoted in such reports are difficult to reconcile or remain questionable, which hampers the acceptance of computational work in interdisciplinary environments. Herein we attempt to highlight selected investigations in which AI has shown promise to impact medicinal chemistry in areas such as compound design and synthesis.