Using simple artificial intelligence methods for predicting amyloidogenesis in antibodies

Cardiac Amyloidosis: A Comprehensive Review of Pathophysiology, Diagnostic Approach, Applications of Artificial Intelligence, and Management Strategies

Cardiac amyloidosis (CA) is a serious and often fatal condition caused by the accumulation of amyloid fibrils in the heart, leading to progressive heart failure. It involves the misfolding of normally soluble proteins into insoluble amyloid fibrils, with transthyretin and light-chain amyloidosis being the most common forms affecting the heart. Advances in diagnostics, especially cardiac magnetic resonance imaging and non-invasive techniques, have improved early detection and disease management. Artificial intelligence has emerged as a diagnostic tool for cardiac amyloidosis, improving accuracy and enabling earlier intervention through advanced imaging analysis and pattern recognition. Management strategies include volume control, specific pharmacotherapies like tafamidis, and addressing arrhythmias and advanced heart failure. However, further research is needed for novel therapeutic approaches, the long-term effectiveness of emerging treatments, and the optimization of artificial intelligence applications in clinical practice for better patient outcomes. The article aims to provide an overview of CA, outlining its pathophysiology, diagnostic advancements, the role of artificial intelligence, management strategies, and the need for further research.

How Artificial Intelligence Can Enhance the Diagnosis of Cardiac Amyloidosis: A Review of Recent Advances and Challenges

Cardiac amyloidosis (CA) is an underdiagnosed form of infiltrative cardiomyopathy caused by abnormal amyloid fibrils deposited extracellularly in the myocardium and cardiac structures. There can be high variability in its clinical manifestations, and diagnosing CA requires expertise and often thorough evaluation; as such, the diagnosis of CA can be challenging and is often delayed. The application of artificial intelligence (AI) to different diagnostic modalities is rapidly expanding and transforming cardiovascular medicine. Advanced AI methods such as deep-learning convolutional neural networks (CNNs) may enhance the diagnostic process for CA by identifying patients at higher risk and potentially expediting the diagnosis of CA. In this review, we summarize the current state of AI applications to different diagnostic modalities used for the evaluation of CA, including their diagnostic and prognostic potential, and current challenges and limitations.

AB-Amy: machine learning aided amyloidogenic risk prediction of therapeutic antibody light chains

Over 120 FDA-approved antibody-based therapeutics are used to treat a variety of diseases.However, many candidates could fail because of unfavorable physicochemical properties. Light-chain amyloidosis is one form of aggregation that can lead to severe safety risks in clinical development. Therefore, screening candidates with a less amyloidosis risk at the early stage can not only save the time and cost of antibody development but also improve the safety of antibody drugs. In this study, based on the dipeptide composition of 742 amyloidogenic and 712 non-amyloidogenic antibody light chains, a support vector machine-based model, AB-Amy, was trained to predict the light-chain amyloidogenic risk. The AUC of AB-Amy reaches 0.9651. The excellent performance of AB-Amy indicates that it can be a useful tool for the in silico evaluation of the light-chain amyloidogenic risk to ensure the safety of antibody therapeutics under clinical development. A web server is freely available at http://i.uestc.edu.cn/AB-Amy/.

Machine Learning Approaches in Diagnosis, Prognosis and Treatment Selection of Cardiac Amyloidosis

Cardiac amyloidosis is an uncommon restrictive cardiomyopathy featuring an unregulated amyloid protein deposition that impairs organic function. Early cardiac amyloidosis diagnosis is generally delayed by indistinguishable clinical findings of more frequent hypertrophic diseases. Furthermore, amyloidosis is divided into various groups, according to a generally accepted taxonomy, based on the proteins that make up the amyloid deposits; a careful differentiation between the various forms of amyloidosis is necessary to undertake an adequate therapeutic treatment. Thus, cardiac amyloidosis is thought to be underdiagnosed, which delays necessary therapeutic procedures, diminishing quality of life and impairing clinical prognosis. The diagnostic work-up for cardiac amyloidosis begins with the identification of clinical features, electrocardiographic and imaging findings suggestive or compatible with cardiac amyloidosis, and often requires the histological demonstration of amyloid deposition. One approach to overcome the difficulty of an early diagnosis is the use of automated diagnostic algorithms. Machine learning enables the automatic extraction of salient information from “raw data” without the need for pre-processing methods based on the a priori knowledge of the human operator. This review attempts to assess the various diagnostic approaches and artificial intelligence computational techniques in the detection of cardiac amyloidosis.

Machine learning prediction of antibody aggregation and viscosity for high concentration formulation development of protein therapeutics

Machine learning has been recently used to predict therapeutic antibody aggregation rates and viscosity at high concentrations (150 mg/ml). These works focused on commercially available antibodies, which may have been optimized for stability. In this study, we measured accelerated aggregation rates at 45°C and viscosity at 150 mg/ml for 20 preclinical and clinical-stage antibodies. Features obtained from molecular dynamics simulations of the full-length antibody and sequences were used for machine learning model construction. We found a k-nearest neighbors regression model with two features, spatial positive charge map on the CDRH2 and solvent-accessible surface area of hydrophobic residues on the variable fragment, gives the best performance for predicting antibody aggregation rates (r = 0.89). For the viscosity classification model, the model with the highest accuracy is a logistic regression model with two features, spatial negative charge map on the heavy chain variable region and spatial negative charge map on the light chain variable region. The accuracy and the area under precision recall curve of the classification model from validation tests are 0.86 and 0.70, respectively. In addition, we combined data from another 27 commercial mAbs to develop a viscosity predictive model. The best model is a logistic regression model with two features, number of hydrophobic residues on the light chain variable region and net charges on the light chain variable region. The accuracy and the area under precision recall curve of the classification model are 0.85 and 0.6, respectively. The aggregation rates and viscosity models can be used to predict antibody stability to facilitate pharmaceutical development.

Exploring the sequence features determining amyloidosis in human antibody light chains

The light chain (AL) amyloidosis is caused by the aggregation of light chain of antibodies into amyloid fibrils. There are plenty of computational resources available for the prediction of short aggregation-prone regions within proteins. However, it is still a challenging task to predict the amyloidogenic nature of the whole protein using sequence/structure information. In the case of antibody light chains, common architecture and known binding sites can provide vital information for the prediction of amyloidogenicity at physiological conditions. Here, in this work, we have compared classical sequence-based, aggregation-related features (such as hydrophobicity, presence of gatekeeper residues, disorderness, β-propensity, etc.) calculated for the CDR, FR or V_L regions of amyloidogenic and non-amyloidogenic antibody light chains and implemented the insights gained in a machine learning-based webserver called "V_LAmY-Pred" ( https://web.iitm.ac.in/bioinfo2/vlamy-pred/ ). The model shows prediction accuracy of 79.7% (sensitivity: 78.7% and specificity: 79.9%) with a ROC value of 0.88 on a dataset of 1828 variable region sequences of the antibody light chains. This model will be helpful towards improved prognosis for patients that may likely suffer from diseases caused by light chain amyloidosis, understanding origins of aggregation in antibody-based biotherapeutics, large-scale in-silico analysis of antibody sequences generated by next generation sequencing, and finally towards rational engineering of aggregation resistant antibodies.

PredAmyl-MLP: Prediction of Amyloid Proteins Using Multilayer Perceptron

Amyloid is generally an aggregate of insoluble fibrin; its abnormal deposition is the pathogenic mechanism of various diseases, such as Alzheimer's disease and type II diabetes. Therefore, accurately identifying amyloid is necessary to understand its role in pathology. We proposed a machine learning-based prediction model called PredAmyl-MLP, which consists of the following three steps: feature extraction, feature selection, and classification. In the step of feature extraction, seven feature extraction algorithms and different combinations of them are investigated, and the combination of SVMProt-188D and tripeptide composition (TPC) is selected according to the experimental results. In the step of feature selection, maximum relevant maximum distance (MRMD) and binomial distribution (BD) are, respectively, used to remove the redundant or noise features, and the appropriate features are selected according to the experimental results. In the step of classification, we employed multilayer perceptron (MLP) to train the prediction model. The 10-fold cross-validation results show that the overall accuracy of PredAmyl-MLP reached 91.59%, and the performance was better than the existing methods.

iAMY-SCM: Improved prediction and analysis of amyloid proteins using a scoring card method with propensity scores of dipeptides

Fast, accurate identification and characterization of amyloid proteins at a large-scale is essential for understating their role in therapeutic intervention strategies. As a matter of fact, there exist only one in silico model for amyloid protein identification using the random forest (RF) model in conjunction with various feature types namely the RFAmy. However, it suffers from low interpretability for biologists. Thus, it is highly desirable to develop a simple and easily interpretable prediction method with robust accuracy as compared to the existing complicated model. In this study, we propose iAMY-SCM, the first scoring card method-based predictor for predicting and analyzing amyloid proteins. Herein, the iAMY-SCM made use of a simple weighted-sum function in conjunction with the propensity scores of dipeptides for the amyloid protein identification. Cross-validation results indicated that iAMY-SCM provided an accuracy of 0.895 that corresponded to 10-22% higher performance than that of widely used machine learning models. Furthermore, iAMY-SCM achieving an accuracy of 0.827 as evaluated by an independent test, which was found to be comparable to that of RFAmy and was approximately 9-13% higher than widely used machine learning models. Furthermore, the analysis of estimated propensity scores of amino acids and dipeptides were performed to provide insights into the biophysical and biochemical properties of amyloid proteins. As such, this demonstrates that the proposed iAMY-SCM is efficient and reliable in terms of simplicity, interpretability and implementation. To facilitate ease of use of the proposed iAMY-SCM, a user-friendly and publicly accessible web server at http://camt.pythonanywhere.com/iAMY-SCM has been established. We anticipate that that iAMY-SCM will be an important tool for facilitating the large-scale prediction and characterization of amyloid protein.

On the Conformational Dynamics of β-Amyloid Forming Peptides: A Computational Perspective

Understanding the conformational dynamics of proteins and peptides involved in important functions is still a difficult task in computational structural biology. Because such conformational transitions in β-amyloid (Aβ) forming peptides play a crucial role in many neurological disorders, researchers from different scientific fields have been trying to address issues related to the folding of Aβ forming peptides together. Many theoretical models have been proposed in the recent years for studying Aβ peptides using mathematical, physicochemical, and molecular dynamics simulation, and machine learning approaches. In this article, we have comprehensively reviewed the developmental advances in the theoretical models for Aβ peptide folding and interactions, particularly in the context of neurological disorders. Furthermore, we have extensively reviewed the advances in molecular dynamics simulation as a tool used for studying the conversions between polymorphic amyloid forms and applications of using machine learning approaches in predicting Aβ peptides and aggregation-prone regions in proteins. We have also provided details on the theoretical advances in the study of Aβ peptides, which would enhance our understanding of these peptides at the molecular level and eventually lead to the development of targeted therapies for certain acute neurological disorders such as Alzheimer's disease in the future.

Identification of amyloidogenic peptides via optimized integrated features space based on physicochemical properties and PSSM

At present, the identification of amyloid becomes more and more essential and meaningful. Because its mis-aggregation may cause some diseases such as Alzheimer's and Parkinson's diseases. This paper focus on the classification of amyloidogenic peptides and a novel feature representation called PhyAve_PSSMDwt is proposed. It includes two parts. One is based on physicochemical properties involving hydrophilicity, hydrophobicity, aggregation tendency, packing density and H-bonding which extracts 15-dimensional features in total. And the other is 60-dimensional features through recursive feature elimination from PSSM by discrete wavelet transform. In this period, sliding window is introduced to reconstruct PSSM so that the evolutionary information of short sequences can still be extracted. At last, the support vector machine is adopted as a classifier. The experimental result on Pep424 dataset shows that PSSM's information makes a great contribution on performance. And compared with other existing methods, our results after cross-validation increase by 3.1%, 3.3%, 0.136 and 0.007 in accuracy, specificity, Matthew's correlation coefficient and AUC value, respectively. It indicates that our method is effective and competitive.

Structure of proteins: Evolution with unsolved mysteries

Evolution of macromolecules could be considered as a milestone in the history of life. Nucleic acids are the long stretches of nucleotides that contain all the possible codes and information of life. On the other hand, proteins are their actual translated outcomes, or reflections of modifications in their structure that have occurred at a slow, but steady rate over a very long period of evolution. Over the years of research, biophysicists, biochemists, molecular and structural biologists have unfurled several layers of the structural convolutions in these chemical molecules; however evolutionists look over their structures through a different prism, which may or may not coincide with others. There remains a need to outline several well-known, but less discussed features of protein structures, like intrinsically disordered states, degron signals and different types of ubiquitin chains providing degradation signals, which help the cellular proteolytic machinery to identify and target the proteins towards degradation pathways. There are several important factors, which are critical for folding of proteins into their native three-dimensional conformations by the cytoplasmic chaperones; but in real time how the chaperones fold the newly synthesized polypeptide sequences into a particular three-dimensional shape within a fraction of second is still a mystery for biologists as well as mathematicians. Multiple similar unsolved or unaddressed questions need to be addressed in detail so that future line of research can dig deeper into the finer details of these structures of the proteins.

RFAmyloid: A Web Server for Predicting Amyloid Proteins

Amyloid is an insoluble fibrous protein and its mis-aggregation can lead to some diseases, such as Alzheimer’s disease and Creutzfeldt–Jakob’s disease. Therefore, the identification of amyloid is essential for the discovery and understanding of disease. We established a novel predictor called RFAmy based on random forest to identify amyloid, and it employed SVMProt 188-D feature extraction method based on protein composition and physicochemical properties and pse-in-one feature extraction method based on amino acid composition, autocorrelation pseudo acid composition, profile-based features and predicted structures features. In the ten-fold cross-validation test, RFAmy’s overall accuracy was 89.19% and F-measure was 0.891. Results were obtained by comparison experiments with other feature, classifiers, and existing methods. This shows the effectiveness of RFAmy in predicting amyloid protein. The RFAmy proposed in this paper can be accessed through the URL http://server.malab.cn/RFAmyloid/.

Breast cancer vaccination comes to age: impacts of bioinformatics

Introduction: Breast cancer, as one of the major causes of cancer death among women, is the central focus of this study. The recent advances in the development and application of computational tools and bioinformatics in the field of immunotherapy of malignancies such as breast cancer have emerged the new dominion of immunoinformatics, and therefore, next generation of immunomedicines .

Methods: Having reviewed the most recent works on the applications of computational tools, we provide comprehensive insights into the breast cancer incidence and its leading causes as well as immunotherapy approaches and the future trends. Furthermore, we discuss the impacts of bioinformatics on different stages of vaccine design for the breast cancer, which can be used to produce much more efficient vaccines through a rationalized time- and cost-effective in silico approaches prior to conducting costly experiments.

Results: The tools can be significantly used for designing the immune system-modulating drugs and vaccines based on in silico approaches prior to in vitro and in vivo experimental evaluations. Application of immunoinformatics in the cancer immunotherapy has shown its success in the pre-clinical models. This success returns back to the impacts of several powerful computational approaches developed during the last decade.

Conclusion: Despite the invention of a number of vaccines for the cancer immunotherapy, more computational and clinical trials are required to design much more efficient vaccines against various malignancies, including breast cancer.

Reconstructing evolutionary trees in parallel for massive sequences

BACKGROUND

Building the evolutionary trees for massive unaligned DNA sequences is challenging and crucial. However, reconstructing evolutionary tree for ultra-large sequences is hard. Massive multiple sequence alignment is also challenging and time/space consuming. Hadoop and Spark are developed recently, which bring spring light for the classical computational biology problems. In this paper, we tried to solve the multiple sequence alignment and evolutionary reconstruction in parallel.

RESULTS

HPTree, which is developed in this paper, can deal with big DNA sequence files quickly. It works well on the >1GB files, and gets better performance than other evolutionary reconstruction tools. Users could use HPTree for reonstructing evolutioanry trees on the computer clusters or cloud platform (eg. Amazon Cloud). HPTree could help on population evolution research and metagenomics analysis.

CONCLUSIONS

In this paper, we employ the Hadoop and Spark platform and design an evolutionary tree reconstruction software tool for unaligned massive DNA sequences. Clustering and multiple sequence alignment are done in parallel. Neighbour-joining model was employed for the evolutionary tree building. We opened our software together with source codes via http://lab.malab.cn/soft/HPtree/ .

Can Simple Interaction Models Explain Sequence-Dependent Effects in Peptide Homodimerization?

The development of rapid methods to explain and predict peptide interactions, aggregation, and self-assembly has become important to understanding amyloid disease pathology, the shelf stability of peptide therapeutics, and the design of novel peptide materials. Although experimental aggregation databases have been used to develop correlative and statistical models, molecular simulations offer atomic-level details that potentially provide greater physical insight and allow one to single out the most explanatory simple models. Here, we outline one such approach using a case study that develops homodimerization models for serine-glycine peptides with various hydrophobic leucine mutations. Using detailed all-atom simulations, we calculate reference dimerization free energy profiles and binding constants for a small peptide library. We then use statistical methods to systematically assess whether simple interaction models, which do not require expensive simulations and free energy calculation, can capture them. Surprisingly, some combinations of a few simple scaling laws well recapitulate the detailed, all-atom results with high accuracy. Specifically, we find that a recently proposed phenomenological hydrophobic force law and coarse measures of entropic effects in binding offer particularly high explanatory power, underscoring the physical relevance to association that these driving forces can play.

From the Evolution of Protein Sequences Able to Resist Self-Assembly to the Prediction of Aggregation Propensity

Folding of polypeptide chains into biologically active entities is an astonishingly complex process, determined by the nature and the sequence of residues emerging from ribosomes. While it has been long believed that evolution has pressed genomes so that specific sequences could adopt unique, functional three-dimensional folds, it is now clear that complex protein machineries act as quality control system and supervise folding. Notwithstanding that, events such as erroneous folding, partial folding, or misfolding are frequent during the life of a cell or a whole organism, and they can escape controls. One of the possible outcomes of this misbehavior is cross-β aggregation, a super secondary structure which represents the hallmark of self-assembled, well organized, and extremely ordered structures termed amyloid fibrils. What if evolution would have not taken into account such possibilities? Twenty years of research point toward the idea that, in fact, evolution has constantly supervised the risk of errors and minimized their impact. In this review we tried to survey the major findings in the amyloid field, trying to describe what the real pitfalls of protein folding are-from an evolutionary perspective-and how sequence and structural features have evolved to balance the need for perfect, dynamic, functionally efficient structures, and the detrimental effects implicit in the dangerous process of folding. We will discuss how the knowledge obtained from these studies has been employed to produce computational methods able to assess, predict, and discriminate the aggregation properties of protein sequences.

FISH Amyloid - a new method for finding amyloidogenic segments in proteins based on site specific co-occurrence of aminoacids

BACKGROUND

Amyloids are proteins capable of forming fibrils whose intramolecular contact sites assume densely packed zipper pattern. Their oligomers can underlie serious diseases, e.g. Alzheimer's and Parkinson's diseases. Recent studies show that short segments of aminoacids can be responsible for amyloidogenic properties of a protein. A few hundreds of such peptides have been experimentally found but experimental testing of all candidates is currently not feasible. Here we propose an original machine learning method for classification of aminoacid sequences, based on discovering a segment with a discriminative pattern of site-specific co-occurrences between sequence elements. The pattern is based on the positions of residues with correlated occurrence over a sliding window of a specified length. The algorithm first recognizes the most relevant training segment in each positive training instance. Then the classification is based on maximal distances between co-occurrence matrix of the relevant segments in positive training sequences and the matrix from negative training segments. The method was applied for studying sequences of aminoacids with regard to their amyloidogenic properties.

RESULTS

Our method was first trained on available datasets of hexapeptides with the amyloidogenic classification, using 5 or 6-residue sliding windows. Depending on the choice of training and testing datasets, the area under ROC curve obtained the value up to 0.80 for experimental, and 0.95 for computationally generated (with 3D profile method) datasets. Importantly, the results on 5-residue segments were not significantly worse, although the classification required that algorithm first recognized the most relevant training segments. The dataset of long sequences, such as sup35 prion and a few other amyloid proteins, were applied to test the method and gave encouraging results. Our web tool FISH Amyloid was trained on all available experimental data 4-10 residues long, offers prediction of amyloidogenic segments in protein sequences.

CONCLUSIONS

We proposed a new original classification method which recognizes co-occurrence patterns in sequences. The method reveals characteristic classification pattern of the data and finds the segments where its scoring is the strongest, also in long training sequences. Applied to the problem of amyloidogenic segments recognition, it showed a good potential for classification problems in bioinformatics.

Antibody Engineering

Advanced molecular biology techniques developed during the past few decades have allowed the industry to exploit and commercialize the natural defense mechanisms that antibodies provide. This review discusses the latest advances in antibody-engineering technologies to enhance clinical efficacy and outcomes. For the constant regions, the choice of the antibody class and isotype has to be made carefully to suit the therapeutic applications. Engineering of the Fc region, either by direct targeted mutagenesis or by modifying the nature of its N -glycan, has played an important role in recent years in increasing half-life or controlling effector functions. The variable regions of the antibody are responsible for binding affinity and exquisite specificity to the target molecule, which together with the Fc determine the drug's efficacy and influence the drug dose required to obtain the desired effectiveness. A key requirement during antibody development is therefore to affinity mature the variable regions when necessary, so that they bind the therapeutic target with sufficiently high affinity to guarantee effective occupancy over prolonged periods. If the antibody was obtained from a non-human source, such as rodents, a humanization process has to be applied to minimize immunogenicity while maintaining the desired binding affinity and selectivity. Finally, we discuss the next next-generation antibodies, such as antibody-drug conjugates, bispecific antibodies, and immunocytokines, which are being developed to meet future challenges.

On the amyloid datasets used for training PAFIG--how (not) to extend the experimental dataset of hexapeptides

BACKGROUND

Amyloids are proteins capable of forming aberrant intramolecular contact sites, characteristic of beta zipper configuration. Amyloids can underlie serious health conditions, e.g. Alzheimer's or Parkinson's diseases. It has been proposed that short segments of amino acids can be responsible for protein amyloidogenicity, but no more than two hundred such hexapeptides have been experimentally found. The authors of the computational tool Pafig published in BMC Bioinformatics a method for extending the amyloid hexapeptide dataset that could be used for training and testing models. They assumed that all hexapeptides belonging to an amyloid protein can be regarded as amylopositive, while those from proteins never reported as amyloid are always amylonegative. Here we show why the above described method of extending datasets is wrong and discuss the reasons why the incorrect data could lead to falsely correct classification.

RESULTS

The amyloid classification of hexapeptides by Pafig was confronted with the classification results from different state of the art computational methods and the outputs of all methods were studied by clustering analysis. The clustering methods show that Pafig is an outlier with regard to other approaches. Our study of the statistical patterns of its training and testing datasets showed a strong bias towards STVIIE hexapeptide in their positive part. Different statistical patterns of seemingly amylo-positive and -negative hexapeptides allow for a repeatable classification, which is not related to amyloid propensity of the hexapetides.

CONCLUSIONS

Our study on recognition of amyloid hexapeptides showed that occurrence of incidental patterns in wrongly selected datasets can produce falsely correct results of classification. The assumption that all hexapeptides belonging to amyloid protein can be regarded as amylopositive and those from proteins never reported as amyloid are always amylonegative is not supported by any other computational method. This is in line with experimental observations that amyloid propensity of a full protein can result from only one amyloidogenic fragment in this protein, while the occurrence of amyliodogenic part that is well hidden inside the protein may never lead to fibril formation. This leads to the conclusion that Pafig does not provide correct classification with regard to amyloidogenicity.

MetAmyl: a METa-predictor for AMYLoid proteins

The aggregation of proteins or peptides in amyloid fibrils is associated with a number of clinical disorders, including Alzheimer's, Huntington's and prion diseases, medullary thyroid cancer, renal and cardiac amyloidosis. Despite extensive studies, the molecular mechanisms underlying the initiation of fibril formation remain largely unknown. Several lines of evidence revealed that short amino-acid segments (hot spots), located in amyloid precursor proteins act as seeds for fibril elongation. Therefore, hot spots are potential targets for diagnostic/therapeutic applications, and a current challenge in bioinformatics is the development of methods to accurately predict hot spots from protein sequences. In this paper, we combined existing methods into a meta-predictor for hot spots prediction, called MetAmyl for METapredictor for AMYLoid proteins. MetAmyl is based on a logistic regression model that aims at weighting predictions from a set of popular algorithms, statistically selected as being the most informative and complementary predictors. We evaluated the performances of MetAmyl through a large scale comparative study based on three independent datasets and thus demonstrated its ability to differentiate between amyloidogenic and non-amyloidogenic polypeptides. Compared to 9 other methods, MetAmyl provides significant improvement in prediction on studied datasets. We further show that MetAmyl is efficient to highlight the effect of point mutations involved in human amyloidosis, so we suggest this program should be a useful complementary tool for the diagnosis of these diseases.

Machine learning methods can replace 3D profile method in classification of amyloidogenic hexapeptides

Background

Amyloids are proteins capable of forming fibrils. Many of them underlie serious diseases, like Alzheimer disease. The number of amyloid-associated diseases is constantly increasing. Recent studies indicate that amyloidogenic properties can be associated with short segments of aminoacids, which transform the structure when exposed. A few hundreds of such peptides have been experimentally found. Experimental testing of all possible aminoacid combinations is currently not feasible. Instead, they can be predicted by computational methods. 3D profile is a physicochemical-based method that has generated the most numerous dataset - ZipperDB. However, it is computationally very demanding. Here, we show that dataset generation can be accelerated. Two methods to increase the classification efficiency of amyloidogenic candidates are presented and tested: simplified 3D profile generation and machine learning methods.

Results

We generated a new dataset of hexapeptides, using more economical 3D profile algorithm, which showed very good classification overlap with ZipperDB (93.5%). The new part of our dataset contains 1779 segments, with 204 classified as amyloidogenic. The dataset of 6-residue sequences with their binary classification, based on the energy of the segment, was applied for training machine learning methods. A separate set of sequences from ZipperDB was used as a test set. The most effective methods were Alternating Decision Tree and Multilayer Perceptron. Both methods obtained area under ROC curve of 0.96, accuracy 91%, true positive rate ca. 78%, and true negative rate 95%. A few other machine learning methods also achieved a good performance. The computational time was reduced from 18-20 CPU-hours (full 3D profile) to 0.5 CPU-hours (simplified 3D profile) to seconds (machine learning).

Conclusions

We showed that the simplified profile generation method does not introduce an error with regard to the original method, while increasing the computational efficiency. Our new dataset proved representative enough to use simple statistical methods for testing the amylogenicity based only on six letter sequences. Statistical machine learning methods such as Alternating Decision Tree and Multilayer Perceptron can replace the energy based classifier, with advantage of very significantly reduced computational time and simplicity to perform the analysis. Additionally, a decision tree provides a set of very easily interpretable rules.

A consensus method for the prediction of 'aggregation-prone' peptides in globular proteins

The purpose of this work was to construct a consensus prediction algorithm of ‘aggregation-prone’ peptides in globular proteins, combining existing tools. This allows comparison of the different algorithms and the production of more objective and accurate results. Eleven (11) individual methods are combined and produce AMYLPRED2, a publicly, freely available web tool to academic users (http://biophysics.biol.uoa.gr/AMYLPRED2), for the consensus prediction of amyloidogenic determinants/‘aggregation-prone’ peptides in proteins, from sequence alone. The performance of AMYLPRED2 indicates that it functions better than individual aggregation-prediction algorithms, as perhaps expected. AMYLPRED2 is a useful tool for identifying amyloid-forming regions in proteins that are associated with several conformational diseases, called amyloidoses, such as Altzheimer's, Parkinson's, prion diseases and type II diabetes. It may also be useful for understanding the properties of protein folding and misfolding and for helping to the control of protein aggregation/solubility in biotechnology (recombinant proteins forming bacterial inclusion bodies) and biotherapeutics (monoclonal antibodies and biopharmaceutical proteins).

Prediction and analysis of antibody amyloidogenesis from sequences

Antibody amyloidogenesis is the aggregation of soluble proteins into amyloid fibrils that is one of major causes of the failures of humanized antibodies. The prediction and prevention of antibody amyloidogenesis are helpful for restoring and enhancing therapeutic effects. Due to a large number of possible germlines, the existing method is not practical to predict sequences of novel germlines, which establishes individual models for each known germline. This study proposes a first automatic and across-germline prediction method (named AbAmyloid) capable of predicting antibody amyloidogenesis from sequences. Since the amyloidogenesis is determined by a whole sequence of an antibody rather than germline-dependent properties such as mutated residues, this study assess three types of germline-independent sequence features (amino acid composition, dipeptide composition and physicochemical properties). AbAmyloid using a Random Forests classifier with dipeptide composition performs well on a data set of 12 germlines. The within- and across-germline prediction accuracies are 83.10% and 83.33% using Jackknife tests, respectively, and the novel-germline prediction accuracy using a leave-one-germline-out test is 72.22%. A thorough analysis of sequence features is conducted to identify informative properties for further providing insights to antibody amyloidogenesis. Some identified informative physicochemical properties are amphiphilicity, hydrophobicity, reverse turn, helical structure, isoelectric point, net charge, mutability, coil, turn, linker, nuclear protein, etc. Additionally, the numbers of ubiquitylation sites in amyloidogenic and non-amyloidogenic antibodies are found to be significantly different. It reveals that antibodies less likely to be ubiquitylated tend to be amyloidogenic. The method AbAmyloid capable of automatically predicting antibody amyloidogenesis of novel germlines is implemented as a publicly available web server at http://iclab.life.nctu.edu.tw/abamyloid.

Computer-aided antibody design

Recent clinical trials using antibodies with low toxicity and high efficiency have raised expectations for the development of next-generation protein therapeutics. However, the process of obtaining therapeutic antibodies remains time consuming and empirical. This review summarizes recent progresses in the field of computer-aided antibody development mainly focusing on antibody modeling, which is divided essentially into two parts: (i) modeling the antigen-binding site, also called the complementarity determining regions (CDRs), and (ii) predicting the relative orientations of the variable heavy (V(H)) and light (V(L)) chains. Among the six CDR loops, the greatest challenge is predicting the conformation of CDR-H3, which is the most important in antigen recognition. Further computational methods could be used in drug development based on crystal structures or homology models, including antibody-antigen dockings and energy calculations with approximate potential functions. These methods should guide experimental studies to improve the affinities and physicochemical properties of antibodies. Finally, several successful examples of in silico structure-based antibody designs are reviewed. We also briefly review structure-based antigen or immunogen design, with application to rational vaccine development.

Categorization of 77 dystrophin exons into 5 groups by a decision tree using indexes of splicing regulatory factors as decision markers

Background

Duchenne muscular dystrophy, a fatal muscle-wasting disease, is characterized by dystrophin deficiency caused by mutations in the dystrophin gene. Skipping of a target dystrophin exon during splicing with antisense oligonucleotides is attracting much attention as the most plausible way to express dystrophin in DMD. Antisense oligonucleotides have been designed against splicing regulatory sequences such as splicing enhancer sequences of target exons. Recently, we reported that a chemical kinase inhibitor specifically enhances the skipping of mutated dystrophin exon 31, indicating the existence of exon-specific splicing regulatory systems. However, the basis for such individual regulatory systems is largely unknown. Here, we categorized the dystrophin exons in terms of their splicing regulatory factors.

Results

Using a computer-based machine learning system, we first constructed a decision tree separating 77 authentic from 14 known cryptic exons using 25 indexes of splicing regulatory factors as decision markers. We evaluated the classification accuracy of a novel cryptic exon (exon 11a) identified in this study. However, the tree mislabeled exon 11a as a true exon. Therefore, we re-constructed the decision tree to separate all 15 cryptic exons. The revised decision tree categorized the 77 authentic exons into five groups. Furthermore, all nine disease-associated novel exons were successfully categorized as exons, validating the decision tree. One group, consisting of 30 exons, was characterized by a high density of exonic splicing enhancer sequences. This suggests that AOs targeting splicing enhancer sequences would efficiently induce skipping of exons belonging to this group.

Conclusions

The decision tree categorized the 77 authentic exons into five groups. Our classification may help to establish the strategy for exon skipping therapy for Duchenne muscular dystrophy.

AGGRESCAN: method, application, and perspectives for drug design

Protein aggregation underlies the development of an increasing number of conformational human diseases of growing incidence, such as Alzheimer's and Parkinson's diseases. Furthermore, the accumulation of recombinant proteins as intracellular aggregates represents a critical obstacle for the biotechnological production of polypeptides. Also, ordered protein aggregates constitute novel and versatile nanobiomaterials. Consequently, there is an increasing interest in the development of methods able to forecast the aggregation properties of polypeptides in order to modulate their intrinsic solubility. In this context, we have developed AGGRESCAN, a simple and fast algorithm that predicts aggregation-prone segments in protein sequences, compares the aggregation properties of different proteins or protein sets and analyses the effect of mutations on protein aggregation propensities.

Dana-Farber repository for machine learning in immunology

The immune system is characterized by high combinatorial complexity that necessitates the use of specialized computational tools for analysis of immunological data. Machine learning (ML) algorithms are used in combination with classical experimentation for the selection of vaccine targets and in computational simulations that reduce the number of necessary experiments. The development of ML algorithms requires standardized data sets, consistent measurement methods, and uniform scales. To bridge the gap between the immunology community and the ML community, we designed a repository for machine learning in immunology named Dana-Farber Repository for Machine Learning in Immunology (DFRMLI). This repository provides standardized data sets of HLA-binding peptides with all binding affinities mapped onto a common scale. It also provides a list of experimentally validated naturally processed T cell epitopes derived from tumor or virus antigens. The DFRMLI data were preprocessed and ensure consistency, comparability, detailed descriptions, and statistically meaningful sample sizes for peptides that bind to various HLA molecules. The repository is accessible at http://bio.dfci.harvard.edu/DFRMLI/.