Machine Learning on DNA-Encoded Library Count Data Using an Uncertainty-Aware Probabilistic Loss Function

QSARtuna: An Automated QSAR Modeling Platform for Molecular Property Prediction in Drug Design

Machine-learning (ML) and deep-learning (DL) approaches to predict the molecular properties of small molecules are increasingly deployed within the design-make-test-analyze (DMTA) drug design cycle to predict molecular properties of interest. Despite this uptake, there are only a few automated packages to aid their development and deployment that also support uncertainty estimation, model explainability, and other key aspects of model usage. This represents a key unmet need within the field, and the large number of molecular representations and algorithms (and associated parameters) means it is nontrivial to robustly optimize, evaluate, reproduce, and deploy models. Here, we present QSARtuna, a molecule property prediction modeling pipeline, written in Python and utilizing the Optuna, Scikit-learn, RDKit, and ChemProp packages, which enables the efficient and automated comparison between molecular representations and machine learning models. The platform was developed by considering the increasingly important aspect of model uncertainty quantification and explainability by design. We provide details for our framework and provide illustrative examples to demonstrate the capability of the software when applied to simple molecular property, reaction/reactivity prediction, and DNA encoded library enrichment classification. We hope that the release of QSARtuna will further spur innovation in automatic ML modeling and provide a platform for education of best practices in molecular property modeling. The code for the QSARtuna framework is made freely available via GitHub.

Machine learning in preclinical drug discovery

Drug-discovery and drug-development endeavors are laborious, costly and time consuming. These programs can take upward of 12 years and cost US $2.5 billion, with a failure rate of more than 90%. Machine learning (ML) presents an opportunity to improve the drug-discovery process. Indeed, with the growing abundance of public and private large-scale biological and chemical datasets, ML techniques are becoming well positioned as useful tools that can augment the traditional drug-development process. In this Perspective, we discuss the integration of algorithmic methods throughout the preclinical phases of drug discovery. Specifically, we highlight an array of ML-based efforts, across diverse disease areas, to accelerate initial hit discovery, mechanism-of-action (MOA) elucidation and chemical property optimization. With advances in the application of ML across diverse therapeutic areas, we posit that fully ML-integrated drug-discovery pipelines will define the future of drug-development programs.

Control of the antitumour activity and specificity of CAR T cells via organic adapters covalently tethering the CAR to tumour cells

On-target off-tumour toxicity limits the anticancer applicability of chimaeric antigen receptor (CAR) T cells. Here we show that the tumour-targeting specificity and activity of T cells with a CAR consisting of an antibody with a lysine residue that catalytically forms a reversible covalent bond with a 1,3-diketone hapten can be regulated by the concentration of a small-molecule adapter. This adapter selectively binds to the hapten and to a chosen tumour antigen via a small-molecule binder identified via a DNA-encoded library. The adapter therefore controls the formation of a covalent bond between the catalytic antibody and the hapten, as well as the tethering of the CAR T cells to the tumour cells, and hence the cytotoxicity and specificity of the cytotoxic T cells, as we show in vitro and in mice with prostate cancer xenografts. Such small-molecule switches of T-cell cytotoxicity and specificity via an antigen-independent 'universal' CAR may enhance the control and safety profile of CAR-based cellular immunotherapies.

Compositional Deep Probabilistic Models of DNA-Encoded Libraries

DNA-encoded library (DEL) has proven to be a powerful tool that utilizes combinatorially constructed small molecules to facilitate highly efficient screening experiments. These selection experiments, involving multiple stages of washing, elution, and identification of potent binders via unique DNA barcodes, often generate complex data. This complexity can potentially mask the underlying signals, necessitating the application of computational tools, such as machine learning, to uncover valuable insights. We introduce a compositional deep probabilistic model of DEL data, DEL-Compose, which decomposes molecular representations into their monosynthon, disynthon, and trisynthon building blocks and capitalizes on the inherent hierarchical structure of these molecules by modeling latent reactions between embedded synthons. Additionally, we investigate methods to improve the observation models for DEL count data, such as integrating covariate factors to more effectively account for data noise. Across two popular public benchmark data sets (CA-IX and HRP), our model demonstrates strong performance compared to count baselines, enriches the correct pharmacophores, and offers valuable insights via its intrinsic interpretable structure, thereby providing a robust tool for the analysis of DEL data.

Evolution of chemistry and selection technology for DNA-encoded library

DNA-encoded chemical library (DEL) links the power of amplifiable genetics and the non-self-replicating chemical phenotypes, generating a diverse chemical world. In analogy with the biological world, the DEL world can evolve by using a chemical central dogma, wherein DNA replicates using the PCR reactions to amplify the genetic codes, DNA sequencing transcripts the genetic information, and DNA-compatible synthesis translates into chemical phenotypes. Importantly, DNA-compatible synthesis is the key to expanding the DEL chemical space. Besides, the evolution-driven selection system pushes the chemicals to evolve under the selective pressure, i.e., desired selection strategies. In this perspective, we summarized recent advances in expanding DEL synthetic toolbox and panning strategies, which will shed light on the drug discovery harnessing in vitro evolution of chemicals via DEL.

DNA-encoded library-enabled discovery of proximity-inducing small molecules

Small molecules that induce protein-protein associations represent powerful tools to modulate cell circuitry. We sought to develop a platform for the direct discovery of compounds able to induce association of any two preselected proteins, using the E3 ligase von Hippel-Lindau (VHL) and bromodomains as test systems. Leveraging the screening power of DNA-encoded libraries (DELs), we synthesized ~1 million DNA-encoded compounds that possess a VHL-targeting ligand, a variety of connectors and a diversity element generated by split-and-pool combinatorial chemistry. By screening our DEL against bromodomains in the presence and absence of VHL, we could identify VHL-bound molecules that simultaneously bind bromodomains. For highly barcode-enriched library members, ternary complex formation leading to bromodomain degradation was confirmed in cells. Furthermore, a ternary complex crystal structure was obtained for our most enriched library member with BRD4BD1 and a VHL complex. Our work provides a foundation for adapting DEL screening to the discovery of proximity-inducing small molecules.

Efficient Computational Framework for Target-Specific Active Peptide Discovery: A Case Study on IL-17C Targeting Cyclic Peptides

The development of potentially active peptides for specific targets is critical for the modern pharmaceutical industry's growth. In this study, we present an efficient computational framework for the discovery of active peptides targeting a specific pharmacological target, which combines a conditional variational autoencoder (CVAE) and a classifier named TCPP based on the Transformer and convolutional neural network. In our example scenario, we constructed an active cyclic peptide library targeting interleukin-17C (IL-17C) through a library-based in vitro selection strategy. The CVAE model is trained on the preprocessed peptide data sets to generate potentially active peptides and the TCPP further screens the generated peptides. Ultimately, six candidate peptides predicted by the model were synthesized and assayed for their activity, and four of them exhibited promising binding affinity to IL-17C. Our study provides a one-stop-shop for target-specific active peptide discovery, which is expected to boost up the process of peptide drug development.

Rational Screening for Cooperativity in Small-Molecule Inducers of Protein-Protein Associations

The hallmark of a molecular glue is its ability to induce cooperative protein-protein interactions, leading to the formation of a ternary complex, despite weaker binding toward one or both individual proteins. Notably, the extent of cooperativity distinguishes molecular glues from bifunctional compounds, which constitute a second class of inducers of protein-protein interactions. However, apart from serendipitous discovery, there have been limited rational screening strategies for the high cooperativity exhibited by molecular glues. Here, we propose a binding-based screen of DNA-barcoded compounds on a target protein in the presence or absence of a presenter protein, using the "presenter ratio", the ratio of ternary enrichment to binary enrichment, as a predictive measure of cooperativity. Through this approach, we identified a range of cooperative, noncooperative, and uncooperative compounds in a single DNA-encoded library screen with bromodomain containing protein (BRD)9 and the VHL-elongin C-elongin B (VCB) complex. Our most cooperative hit compound, 13-7, exhibits micromolar binding affinity to BRD9 but nanomolar affinity for the ternary complex with BRD9 and VCB, with cooperativity comparable to classical molecular glues. This approach may enable the rational discovery of molecular glues for preselected proteins and thus facilitate the transition to a new paradigm of small-molecule therapeutics.

MBE: model-based enrichment estimation and prediction for differential sequencing data

Characterizing differences in sequences between two conditions, such as with and without drug exposure, using high-throughput sequencing data is a prevalent problem involving quantifying changes in sequence abundances, and predicting such differences for unobserved sequences. A key shortcoming of current approaches is their extremely limited ability to share information across related but non-identical reads. Consequently, they cannot use sequencing data effectively, nor be directly applied in many settings of interest. We introduce model-based enrichment (MBE) to overcome this shortcoming. We evaluate MBE using both simulated and real data. Overall, MBE improves accuracy compared to current differential analysis methods.

Building Block-Based Binding Predictions for DNA-Encoded Libraries

DNA-encoded libraries (DELs) provide the means to make and screen millions of diverse compounds against a target of interest in a single experiment. However, despite producing large volumes of binding data at a relatively low cost, the DEL selection process is susceptible to noise, necessitating computational follow-up to increase signal-to-noise ratios. In this work, we present a set of informatics tools to employ data from prior DEL screen(s) to gain information about which building blocks are most likely to be productive when designing new DELs for the same target. We demonstrate that similar building blocks have similar probabilities of forming compounds that bind. We then build a model from the inference that the combined behavior of individual building blocks is predictive of whether an overall compound binds. We illustrate our approach on a set of three-cycle OpenDEL libraries screened against soluble epoxide hydrolase (sEH) and report performance of more than an order of magnitude greater than random guessing on a holdout set, demonstrating that our model can serve as a baseline for comparison against other machine learning models on DEL data. Lastly, we provide a discussion on how we believe this informatics workflow could be applied to benefit researchers in their specific DEL campaigns.

PepScaf: Harnessing Machine Learning with In Vitro Selection toward De Novo Macrocyclic Peptides against IL-17C/IL-17RE Interaction

The combination of library-based screening and artificial intelligence (AI) has been accelerating the discovery and optimization of hit ligands. However, the potential of AI to assist in de novo macrocyclic peptide ligand discovery has yet to be fully explored. In this study, an integrated AI framework called PepScaf was developed to extract the critical scaffold relative to bioactivity based on a vast dataset from an initial in vitro selection campaign against a model protein target, interleukin-17C (IL-17C). Taking the generated scaffold, a focused macrocyclic peptide library was rationally constructed to target IL-17C, yielding over 20 potent peptides that effectively inhibited IL-17C/IL-17RE interaction. Notably, the top two peptides displayed exceptional potency with IC50 values of 1.4 nM. This approach presents a viable methodology for more efficient macrocyclic peptide discovery, offering potential time and cost savings. Additionally, this is also the first report regarding the discovery of macrocyclic peptides against IL-17C/IL-17RE interaction.

Deep Learning Approach for the Discovery of Tumor-Targeting Small Organic Ligands from DNA-Encoded Chemical Libraries

DNA-Encoded Chemical Libraries (DELs) have emerged as efficient and cost-effective ligand discovery tools, which enable the generation of protein-ligand interaction data of unprecedented size. In this article, we present an approach that combines DEL screening and instance-level deep learning modeling to identify tumor-targeting ligands against carbonic anhydrase IX (CAIX), a clinically validated marker of hypoxia and clear cell renal cell carcinoma. We present a new ligand identification and hit-to-lead strategy driven by machine learning models trained on DELs, which expand the scope of DEL-derived chemical motifs. CAIX-screening datasets obtained from three different DELs were used to train machine learning models for generating novel hits, dissimilar to elements present in the original DELs. Out of the 152 novel potential hits that were identified with our approach and screened in an in vitro enzymatic inhibition assay, 70% displayed submicromolar activities (IC₅₀ < 1 μM). To generate lead compounds that are functionalized with anticancer payloads, analogues of top hits were prioritized for synthesis based on the predicted CAIX affinity and synthetic feasibility. Three lead candidates showed accumulation on the surface of CAIX-expressing tumor cells in cellular binding assays. The best compound displayed an in vitro K_D of 5.7 nM and selectively targeted tumors in mice bearing human renal cell carcinoma lesions. Our results demonstrate the synergy between DEL and machine learning for the identification of novel hits and for the successful translation of lead candidates for in vivo targeting applications.

Machine-Learning-Based Data Analysis Method for Cell-Based Selection of DNA-Encoded Libraries

DNA-encoded library (DEL) is a powerful ligand discovery technology that has been widely adopted in the pharmaceutical industry. DEL selections are typically performed with a purified protein target immobilized on a matrix or in solution phase. Recently, DELs have also been used to interrogate the targets in the complex biological environment, such as membrane proteins on live cells. However, due to the complex landscape of the cell surface, the selection inevitably involves significant nonspecific interactions, and the selection data are much noisier than the ones with purified proteins, making reliable hit identification highly challenging. Researchers have developed several approaches to denoise DEL datasets, but it remains unclear whether they are suitable for cell-based DEL selections. Here, we report the proof-of-principle of a new machine-learning (ML)-based approach to process cell-based DEL selection datasets by using a Maximum A Posteriori (MAP) estimation loss function, a probabilistic framework that can account for and quantify uncertainties of noisy data. We applied the approach to a DEL selection dataset, where a library of 7,721,415 compounds was selected against a purified carbonic anhydrase 2 (CA-2) and a cell line expressing the membrane protein carbonic anhydrase 12 (CA-12). The extended-connectivity fingerprint (ECFP)-based regression model using the MAP loss function was able to identify true binders and also reliable structure-activity relationship (SAR) from the noisy cell-based selection datasets. In addition, the regularized enrichment metric (known as MAP enrichment) could also be calculated directly without involving the specific machine-learning model, effectively suppressing low-confidence outliers and enhancing the signal-to-noise ratio. Future applications of this method will focus on de novo ligand discovery from cell-based DEL selections.

Rational screening for cooperativity in small-molecule inducers of protein-protein associations

The hallmark of a molecular glue is its ability to induce cooperative protein-protein interactions, leading to the formation of a ternary complex, despite weaker binding towards one or both individual proteins. Notably, the extent of cooperativity distinguishes molecular glues from bifunctional compounds, a second class of inducers of protein-protein interactions. However, apart from serendipitous discovery, there have been limited rational screening strategies for the high cooperativity exhibited by molecular glues. Here, we propose a binding-based screen of DNA-barcoded compounds on a target protein in the presence and absence of a presenter protein, using the "presenter ratio", the ratio of ternary enrichment to binary enrichment, as a predictive measure of cooperativity. Through this approach, we identified a range of cooperative, noncooperative, and uncooperative compounds in a single DNA-encoded library screen with bromodomain (BRD)9 and the VHL-elongin C-elongin B (VCB) complex. Our most cooperative hit compound, 13-7 , exhibits micromolar binding affinity to BRD9 but nanomolar affinity for the ternary complex with BRD9 and VCB, with cooperativity comparable to classical molecular glues. This approach may enable the discovery of molecular glues for pre-selected proteins and thus facilitate the transition to a new paradigm of molecular therapeutics.

Translating the Genome into Drugs

The Human Genome Project ultimately aimed to translate DNA sequence into drugs. With the draft in hand, the Molecular Libraries Program set out to prosecute all genome-encoded proteins for drug discovery with automated high-throughput screening (HTS). This ambitious vision remains unfulfilled, even while innovations in sequencing technology have fully democratized access to genome-scale sequencing. Why? While the central dogma of biology allows us to chart the entirety of cellular metabolism through sequencing, there is no direct coding for chemistry. The rules of base pairing that relate DNA gene to RNA transcript and amino acid sequence do not exist for relating small-molecule structure with macromolecular binding partners and subsequently cellular function. Obtaining such relationships genome-wide is unapproachable via state-of-the-art HTS, akin to attempting genome-wide association studies using turn-of-the-millennium Sanger DNA sequencing.Our laboratory has been engaged in a multipronged technology development campaign to revolutionize molecular screening through miniaturization in pursuit of genome-scale drug discovery capabilities. The compound library was ripe for miniaturization: it clearly needed to become a consumable. We employed DNA-encoded library (DEL) synthesis principles in the development of solid-phase DELs prepared on microscopic beads, each harboring 100 fmol of a single library member and a DNA tag whose sequence describes the structure of the library member. Loading these DEL beads into 100 pL microfluidic droplets followed by online photocleavage, incubation, fluorescence-activated droplet sorting, and DNA sequencing of the sorted DEL beads reveals the chemical structures of bioactive compounds. This scalable library synthesis and screening platform has proven useful in several proof-of-concept projects involving current clinical targets.Moving forward, we face the problem of druggability and proteome-scale assay development. Developing biochemical or cellular assays for all genome-encoded targets is not scalable and likely impossible as most proteins have ill-defined or unknown activity and may not function outside of their native contexts. These are the dark undruggable expanses, and charting them will require advanced synthesis and analytical technologies that can generalize probe discovery, irrespective of mature protein function, to fulfill the Genome Project's vision of proteome-wide control of cellular pharmacology.

Growing ecosystem of deep learning methods for modeling protein-protein interactions

Numerous cellular functions rely on protein-protein interactions. Efforts to comprehensively characterize them remain challenged however by the diversity of molecular recognition mechanisms employed within the proteome. Deep learning has emerged as a promising approach for tackling this problem by exploiting both experimental data and basic biophysical knowledge about protein interactions. Here, we review the growing ecosystem of deep learning methods for modeling protein interactions, highlighting the diversity of these biophysically informed models and their respective trade-offs. We discuss recent successes in using representation learning to capture complex features pertinent to predicting protein interactions and interaction sites, geometric deep learning to reason over protein structures and predict complex structures, and generative modeling to design de novo protein assemblies. We also outline some of the outstanding challenges and promising new directions. Opportunities abound to discover novel interactions, elucidate their physical mechanisms, and engineer binders to modulate their functions using deep learning and, ultimately, unravel how protein interactions orchestrate complex cellular behaviors.

Discovery of TIGIT inhibitors based on DEL and machine learning

Drug discovery has entered a new period of vigorous development with advanced technologies such as DNA-encoded library (DEL) and artificial intelligence (AI). The previous DEL-AI combination has been successfully applied in the drug discovery of classical kinase and receptor targets mainly based on the known scaffold. So far, there is no report of the DEL-AI combination on inhibitors targeting protein-protein interaction, including those undruggable targets with few or unknown active scaffolds. Here, we applied DEL technology on the T cell immunoglobulin and ITIM domain (TIGIT) target, resulting in the unique hit compound 1 (IC50 = 20.7 μM). Based on the screening data from DEL and hit derivatives a1-a34, a machine learning (ML) modeling process was established to address the challenge of poor sample distribution uniformity, which is also frequently encountered in DEL screening on new targets. In the end, the established ML model achieved a satisfactory hit rate of about 75% for derivatives in a high-scored area.

Cancer Therapeutic Targeting of Hypoxia Induced Carbonic Anhydrase IX: From Bench to Bedside

Simple Summary

Tumor hypoxia remains a significant problem in the effective treatment of most cancers. Tumor cells within hypoxic niches tend to be largely resistant to most therapeutic modalities, and adaptation of the cells within the hypoxic microenvironment imparts the cells with aggressive, invasive behavior. Thus, a major goal of successful cancer therapy should be the eradication of hypoxic tumor cells. Carbonic Anhydrase IX (CAIX) is an exquisitely hypoxia induced protein, selectively expressed on hypoxic tumor cells, and thus has garnered significant attention as a therapeutic target. In this Commentary, we discuss the current status of targeting CAIX, and future strategies for effective, durable cancer treatment.

Abstract

Carbonic Anhydrase IX (CAIX) is a major metabolic effector of tumor hypoxia and regulates intra- and extracellular pH and acidosis. Significant advances have been made recently in the development of therapeutic targeting of CAIX. These approaches include antibody-based immunotherapy, as well as use of antibodies to deliver toxic and radioactive payloads. In addition, a large number of small molecule inhibitors which inhibit the enzymatic activity of CAIX have been described. In this commentary, we highlight the current status of strategies targeting CAIX in both the pre-clinical and clinical space, and discuss future perspectives that leverage inhibition of CAIX in combination with additional targeted therapies to enable effective, durable approaches for cancer therapy.