Sequence-based Computational Method Research Articles

Deep-STP: a deep learning-based approach to predict snake toxin proteins by using word embeddings.

Snake venom contains many toxic proteins that can destroy the circulatory system or nervous system of prey. Studies have found that these snake venom proteins have the potential to treat cardiovascular and nervous system diseases. Therefore, the study of snake venom protein is conducive to the development of related drugs. The research technologies based on traditional biochemistry can accurately identify these proteins, but the experimental cost is high and the time is long. Artificial intelligence technology provides a new means and strategy for large-scale screening of snake venom proteins from the perspective of computing. In this paper, we developed a sequence-based computational method to recognize snake toxin proteins. Specially, we utilized three different feature descriptors, namely g-gap, natural vector and word 2 vector, to encode snake toxin protein sequences. The analysis of variance (ANOVA), gradient-boost decision tree algorithm (GBDT) combined with incremental feature selection (IFS) were used to optimize the features, and then the optimized features were input into the deep learning model for model training. The results show that our model can achieve a prediction performance with an accuracy of 82.00% in 10-fold cross-validation. The model is further verified on independent data, and the accuracy rate reaches to 81.14%, which demonstrated that our model has excellent prediction performance and robustness.

Structure-free antibody paratope similarity prediction for in silico epitope binning via protein language models

Antibodies are an important group of biological molecules that are used as therapeutics and diagnostic tools. Although millions of antibody sequences are available, identifying their structural and functional similarity and their antigen binding sites remains a challenge at large scale. Here, we present a fast, sequence-based computational method for antibody paratope prediction based on protein language models. The paratope information is then used to measure similarity among antibodies via protein language models. Our computational method enables binning of antibody discovery hits into groups as the function of epitope engagement. We further demonstrate the utility of the method by identifying antibodies targeting highly similar epitopes of the same antigens from a large pool of antibody sequences, using two case studies: SARS CoV2 Receptor Binding Domain (RBD) and Epidermal Growth Factor Receptor (EGFR). Our approach highlights the potential in accelerating antibody discovery by enhancing hit prioritization and diversity selection.

CLNN-loop: a deep learning model to predict CTCF-mediated chromatin loops in the different cell lines and CTCF-binding sites (CBS) pair types.

Three-dimensional (3D) genome organization is of vital importance in gene regulation and disease mechanisms. Previous studies have shown that CTCF-mediated chromatin loops are crucial to studying the 3D structure of cells. Although various experimental techniques have been developed to detect chromatin loops, they have been found to be time-consuming and costly. Nowadays, various sequence-based computational methods can capture significant features of 3D genome organization and help predict chromatin loops. However, these methods have low performance and poor generalization ability in predicting chromatin loops. Here, we propose a novel deep learning model, called CLNN-loop, to predict chromatin loops in different cell lines and CTCF-binding sites (CBS) pair types by fusing multiple sequence-based features. The analysis of a series of examinations based on the datasets in the previous study shows that CLNN-loop has satisfactory performance and is superior to the existing methods in terms of predicting chromatin loops. In addition, we apply the SHAP framework to interpret the predictions of different models, and find that CTCF motif and sequence conservation are important signs of chromatin loops in different cell lines and CBS pair types. The source code of CLNN-loop is freely available at https://github.com/HaoWuLab-Bioinformatics/CLNN-loop and the webserver of CLNN-loop is freely available at http://hwclnn.sdu.edu.cn. Supplementary data are available at Bioinformatics online.

IDRBP-ECHF: Identifying DNA- and RNA-binding proteins based on extensible cubic hybrid framework

Proteins interact with nucleic acids to regulate the life activities of organisms. Therefore, how to accurately and efficiently identify nucleic acid-binding proteins (NABPs) is particularly significant. Some sequence-based computational methods have been proposed to identify DNA- and RNA-binding proteins in previous studies. However, the benchmark datasets used by these methods ignore the proportion of NABPs in the real world, and some integration methods only integrate traditional machine learning algorithms, resulting in limited prediction performance. In this study, we proposed a sequence-based method called iDRBP-ECHF to predict the DNA-binding proteins (DBPs) and RNA-binding proteins (RBPs). We constructed a benchmark dataset by considering the proportion of positive and negative samples in the real world, and used down-sampling to generate three relatively balanced datasets to train the iDRBP-ECHF. In addition, we incorporated the deep learning algorithms into the framework to obtain a more compact high-level feature representation of the input data. The results on two independent datasets show that it achieves the most advanced performance and is superior to the other existing sequence-based DBP and RBP prediction methods. In addition, we set up a webserver iDRBP-ECHF, which can be accessed at http://bliulab.net/iDRBP-ECHF.

Immunoinformatic screening of Marburgvirus epitopes and computational investigations of epitope-allele complexes

Marburgvirus (MARV), a member of the Filovirus family, causes severe hemorrhagic fever in humans. Currently, there are no approved vaccines or post exposure treatment methods available against MARV. With the aim of identifying vaccine candidates against MARV, we employ different sequence-based computational methods to predict the MHC-I and MHC-II T-cell epitopes as well as B-cell epitopes for the complete MARV genome. We analyzed the variations in the predicted epitopes among four MARV variants, the Lake Victoria, Angola, Musoke, and Ravn. We used a consensus approach to identify several epitopes, including novel epitopes, and narrowed down the selection based on different parameters such as antigenicity and IC50 values. The selected epitopes can be used in various vaccine constructs that give effective antibody responses. The MHC-I epitope-allele complexes for GP and NP with favorably low IC50 values were investigated using molecular dynamics computations to determine the molecular details of the epitope-allele complexes. This study provides information for further experimental validation of the potential epitopes and the design and development of MARV vaccines.

IDRNA-ITF: identifying DNA- and RNA-binding residues in proteins based on induction and transfer framework.

Protein-DNA and protein-RNA interactions are involved in many biological activities. In the post-genome era, accurate identification of DNA- and RNA-binding residues in protein sequences is of great significance for studying protein functions and promoting new drug design and development. Therefore, some sequence-based computational methods have been proposed for identifying DNA- and RNA-binding residues. However, they failed to fully utilize the functional properties of residues, leading to limited prediction performance. In this paper, a sequence-based method iDRNA-ITF was proposed to incorporate the functional properties in residue representation by using an induction and transfer framework. The properties of nucleic acid-binding residues were induced by the nucleic acid-binding residue feature extraction network, and then transferred into the feature integration modules of the DNA-binding residue prediction network and the RNA-binding residue prediction network for the final prediction. Experimental results on four test sets demonstrate that iDRNA-ITF achieves the state-of-the-art performance, outperforming the other existing sequence-based methods. The webserver of iDRNA-ITF is freely available at http://bliulab.net/iDRNA-ITF.

Multiple Profile Models Extract Features from Protein Sequence Data and Resolve Functional Diversity of Very Different Protein Families.

Functional classification of proteins from sequences alone has become a critical bottleneck in understanding the myriad of protein sequences that accumulate in our databases. The great diversity of homologous sequences hides, in many cases, a variety of functional activities that cannot be anticipated. Their identification appears critical for a fundamental understanding of the evolution of living organisms and for biotechnological applications. ProfileView is a sequence-based computational method, designed to functionally classify sets of homologous sequences. It relies on two main ideas: the use of multiple profile models whose construction explores evolutionary information in available databases, and a novel definition of a representation space in which to analyze sequences with multiple profile models combined together. ProfileView classifies protein families by enriching known functional groups with new sequences and discovering new groups and subgroups. We validate ProfileView on seven classes of widespread proteins involved in the interaction with nucleic acids, amino acids and small molecules, and in a large variety of functions and enzymatic reactions. ProfileView agrees with the large set of functional data collected for these proteins from the literature regarding the organization into functional subgroups and residues that characterize the functions. In addition, ProfileView resolves undefined functional classifications and extracts the molecular determinants underlying protein functional diversity, showing its potential to select sequences towards accurate experimental design and discovery of novel biological functions. On protein families with complex domain architecture, ProfileView functional classification reconciles domain combinations, unlike phylogenetic reconstruction. ProfileView proves to outperform the functional classification approach PANTHER, the two k-mer-based methods CUPP and eCAMI and a neural network approach based on Restricted Boltzmann Machines. It overcomes time complexity limitations of the latter.

DiSBPred: A machine learning based approach for disulfide bond prediction

The protein disulfide bond is a covalent bond that forms during post-translational modification by the oxidation of a pair of cysteines. In protein, the disulfide bond is the most frequent covalent link between amino acids after the peptide bond. It plays a significant role in three-dimensional (3D) ab initio protein structure prediction (aiPSP), stabilizing protein conformation, post-translational modification, and protein folding. In aiPSP, the location of disulfide bonds can strongly reduce the conformational space searching by imposing geometrical constraints. Existing experimental techniques for the determination of disulfide bonds are time-consuming and expensive. Thus, developing sequence-based computational methods for disulfide bond prediction becomes indispensable. This study proposed a stacking-based machine learning approach for disulfide bond prediction (diSBPred). Various useful sequence and structure-based features are extracted for effective training, including conservation profile, residue solvent accessibility, torsion angle flexibility, disorder probability, a sequential distance between cysteines, and more. The prediction of disulfide bonds is carried out in two stages: first, individual cysteines are predicted as either bonding or non-bonding; second, the cysteine-pairs are predicted as either bonding or non-bonding by including the results from cysteine bonding prediction as a feature.The examination of the relevance of the features employed in this study and the features utilized in the existing nearest neighbor algorithm (NNA) method shows that the features used in this study improve about 7.39 % in jackknife validation balanced accuracy. Moreover, for individual cysteine bonding prediction and cysteine-pair bonding prediction, diSBPred provides a 10-fold cross-validation balanced accuracy of 82.29 % and 94.20 %, respectively. Altogether, our predictor achieves an improvement of 43.25 % based on balanced accuracy compared to the existing NNA based approach. Thus, diSBPred can be utilized to annotate the cysteine bonding residues of protein sequences whose structures are unknown as well as improve the accuracy of the aiPSP method, which can further aid in experimental studies of the disulfide bond and structure determination.

IDRBP_MMC: Identifying DNA-Binding Proteins and RNA-Binding Proteins Based on Multi-Label Learning Model and Motif-Based Convolutional Neural Network

DNA-binding protein (DBP) and RNA-binding protein (RBP) are playing crucial roles in gene expression. Accurate identification of them is of great significance, and accurately computational predictors are highly required. In previous studies, DBP recognition and RBP recognition were treated as two separate tasks. Because the functional and structural similarities between DBPs and RBPs are high, the DBP predictors tend to predict RBPs as DBPs, while the RBP predictors tend to predict the DBPs as the RBPs, leading to high cross-prediction rate and low prediction precision. Here we introduced a multi-label learning model based on the motif-based convolutional neural network, and a sequence-based computational method called iDRBP_MMC was proposed to solve the cross-prediction problem so as to improve the predictive performance of DBPs and RBPs. The results on four test datasets showed that it outperformed other state-of-the-art DBP predictors and RBP predictors. When applied to analyze the tomato genome, the results reveal the ability of iDRBP_MMC for large-scale data analysis. Moreover, iDRBP_MMC can identify the proteins binding to both DNA and RNA, which is beyond the scope of existing DBP predictors or RBP predictors. The web-server of iDRBP_MMC is freely available at http://bliulab.net/iDRBP_MMC.

Automated recommendation of the seasonal influenza vaccine strain with PREDAC

Vaccination is currently the most effective way to protect against influenza viruses. This study presents PREDAC (PREDict Antigenic Cluster), a sequence-based computational method to model antigenic clusters of influenza viruses. PREDAC could improve vaccine recommendations for influenza viruses and has been used to support the influenza vaccine strain recommendations at the Chinese National Influenza Center. Moreover, a PREDAC server which includes tools for the prediction of antigenic variants and antigenic clusters of both influenza A and B viruses was built and made publicly available at http://www.computationalbiology.cn/home/. PREDAC could not only facilitate systematic studies of the antigenic evolution of influenza viruses but also provide user-friendly tools for vaccine strain recommendations for influenza viruses.

Pred-MutHTP: Prediction of disease-causing and neutral mutations in human transmembrane proteins.

Membrane proteins are unique in that segments thereof concurrently reside in vastly different physicochemical environments: the extracellular space, the lipid bilayer, and the cytoplasm. Accordingly, the effects of missense variants disrupting their sequence depend greatly on the characteristics of the environment of the protein segment affected as well as the function it performs. Because membrane proteins have many crucial roles (transport, signal transduction, cell adhesion, etc.), compromising their functionality often leads to diseases including cancers, diabetes mellitus or cystic fibrosis. Here, we report a suite of sequence-based computational methods "Pred-MutHTP" for discriminating between disease-causing and neutral alterations in their sequence. With a data set of 11,846 disease-causing and 9,533 neutral mutations, we obtained an accuracy of 74% and 78% with 10-fold group-wise cross-validation and test set, respectively. The features used in the models include evolutionary information, physiochemical properties, neighboring residue information, and specialized membrane protein attributes incorporating the number of transmembrane segments, substitution matrices specific to membrane proteins as well as residue distributions occurring in specific topological regions. Across 11 disease classes, the method achieved accuracies in the range of 75-85%. The model designed specifically for the transmembrane segments achieved an accuracy of 85% on the test set with a sensitivity and specificity of 86% and 83%, respectively. This renders our method the current state-of-the-art with regard to predicting the effects of variants in the transmembrane protein segments. Pred-MutHTP allows predicting the effect of any variant occurring in a membrane protein-available at https://www.iitm.ac.in/bioinfo/PredMutHTP/.

Improved Prediction of Protein-Protein Interactions Using Descriptors Derived From PSSM via Gray Level Co-Occurrence Matrix

A better exploring biological processes, means, and functions demands trusted information about Protein-protein interactions (PPIs). High-throughput technologies have produced a large number of PPIs data for various species, however, they are resource-expensive and often suffer from high error rates. To supplement the limitations of the traditional methods, in this paper, a sequence-based computational method is proposed to insight whether two proteins interact or not. The proposed method divides the novel PPIs prediction process into three stages: first, the position-specific scoring matrices (PSSMs) are produced by incorporating the evolutionary information; second, the 352-dimensional feature vector is constructed for each protein pair; third, effective parameters for the ensemble learning algorithm rotation forest (RF) are selected. In the proposed model, the evolutionary features are extracted from PSSM for each protein without considering any protein annotations. In addition, by using more accurate and diverse classifiers constructed by RF algorithm to avoid yielding coincident errors, one sample incorrectly divided by one classifier will be corrected by another classifier. The proposed method is evaluated in terms of accuracy, precision, sensitivity, and so on using Yeast, Human, and Pylori datasets and finds that its performance is superior to that of the competing methods. Specifically, the average accuracies achieved by the proposed method are 97.06% (Yeast), 98.95% (Human), and 89.69% (H.pylori), which improves the accuracy of PPIs prediction by 0.54%similar to 3.89% (Yeast), 1.29%similar to 3.85% (Human), and 0.22%similar to 4.85% (H.pylori). The experimental results prove that the proposed method is an effective alternative approach for predicting novel PPIs.

IPhosY-PseAAC: identify phosphotyrosine sites by incorporating sequence statistical moments into PseAAC.

Protein phosphorylation is one of the most fundamental types of post-translational modifications and it plays a vital role in various cellular processes of eukaryotes. Among three types of phosphorylation i.e. serine, threonine and tyrosine phosphorylation, tyrosine phosphorylation is one of the most frequent and it is important for mediation of signal transduction in eukaryotic cells. Site-directed mutagenesis and mass spectrometry help in the experimental determination of cellular signalling networks, however, these techniques are costly, time taking and labour associated. Thus, efficient and accurate prediction of these sites through computational approaches can be beneficial to reduce cost and time. Here, we present a more accurate and efficient sequence-based computational method for prediction of phosphotyrosine (PhosY) sites by incorporation of statistical moments into PseAAC. The study is carried out based on Chou's 5-step rule, and various position-composition relative features are used to train a neural network for the prediction purpose. Validation of results through Jackknife testing is performed to validate the results of the proposed prediction method. Overall accuracy validated through Jackknife testing was calculated 93.9%. These results suggest that the proposed prediction model can play a fundamental role in the prediction of PhosY sites in an accurate and efficient way.

MemBrain: An Easy-to-Use Online Webserver for Transmembrane Protein Structure Prediction

Membrane proteins are an important kind of proteins embedded in the membranes of cells and play crucial roles in living organisms, such as ion channels, transporters, receptors. Because it is difficult to determinate the membrane protein’s structure by wet-lab experiments, accurate and fast amino acid sequence-based computational methods are highly desired. In this paper, we report an online prediction tool called MemBrain, whose input is the amino acid sequence. MemBrain consists of specialized modules for predicting transmembrane helices, residue–residue contacts and relative accessible surface area of α-helical membrane proteins. MemBrain achieves a prediction accuracy of 97.9% of ATMH, 87.1% of AP, 3.2 ± 3.0 of N-score, 3.1 ± 2.8 of C-score. MemBrain-Contact obtains 62%/64.1% prediction accuracy on training and independent dataset on top L/5 contact prediction, respectively. And MemBrain-Rasa achieves Pearson correlation coefficient of 0.733 and its mean absolute error of 13.593. These prediction results provide valuable hints for revealing the structure and function of membrane proteins. MemBrain web server is free for academic use and available at www.csbio.sjtu.edu.cn/bioinf/MemBrain/.

A sequence-based computational method for prediction of MoRFs

Molecular recognition features (MoRFs) are relatively short segments (10–70 residues) within intrinsically disordered regions (IDRs) that can undergo disorder-to-order transitions during binding to partner proteins.

IFrag: A Protein–Protein Interface Prediction Server Based on Sequence Fragments

Protein–protein interactions (PPIs) are crucial in many biological processes. The first step towards the molecular characterisation of PPIs implies the charting of their interfaces, that is, the surfaces mediating the interaction. To this end, we present here iFrag, a sequence-based computational method that infers possible interacting regions between two proteins by searching minimal common sequence fragments of the interacting protein pairs. By utilising the sequences of two interacting proteins (queries), iFrag derives a two-dimensional matrix computing a score for each pair of residues that relates to the presence of similar regions in interolog protein pairs. The scoring matrix is represented as a heat map reflecting the potential interface regions in both query proteins. Unlike existing approaches, iFrag does not require three-dimensional structural information or multiple sequence alignments and can even predict small interaction sites consisting only of few residues. Thus, predicted interfaces range from short fragments composed of few residues to domains of proteins, depending on available information on PPIs, as we demonstrate in several examples. Moreover, as a proof of concept, we include the experimental validation on the successful prediction of a peptide competing with the aggregation of β-amyloid in Alzheimer's disease. iFrag is freely accessible at http://sbi.imim.es/iFrag.

Abstract LB-287: Combining immunomics and genomics for immunotherapy of refractory and rare cancers

Abstract Introduction: An underlying basis of cancer immunotherapy is the activation of the patient's own immune effector cells to specifically eradicate tumor cells. Different strategies of cancer immunotherapy have shown outstanding success in a number of patients across various tumor types. However, currently only a fraction of patients respond to immunotherapy, and therapy works only in a subset of tumors. To understand why, an integrated analysis of host and tumor factors is required. Fundamental issues in the analysis of key prognostic and predictive components include elucidating the complex interaction between tumors and their microenvironment, determining the type and densities of tumor-infiltrating immune cells, and understanding how the mutational load influences therapy outcome. Sophisticated analysis is required for patients with refractory or rare cancers to explore new therapeutic strategies. We established a systems medicine approach to obtain detailed data about tumor-host interactions, which may provide information on clinical outcome and help identify patients who are most likely to benefit from immunotherapy. Methods and Results: Whole-exome sequencing of tumor and control samples is performed from freshly obtained patient samples. In addition, whole-transcriptome sequencing of tumor tissue is performed if possible. Single nucleotide variations (SNVs) are detected in each tumor. Sequence-based HLA typing and computational epitope prediction methods are used to assess the immunogenicity of each SNV. The grade of immune cell infiltration and the expression of selected immunomodulatory proteins, e.g. PD-1 and PD-L1, is assessed by immunohistochemical stainings of tissue samples. Additionally, cytokine and chemokine profiles are determined in the tumor microenvironment. The resulting data can aid in treatment decision, e.g. PD-1 blocking antibodies can be administered when elevated PD-1 expression is detected. Alternatively, peptide vaccination is considered if a sufficient number of immunogenic mutations is detected. Based on the mutational profiles, mutation specific 29mer peptides and corresponding wild type peptides are produced for ELIspot assays with patient PBMCs to assess mutation-specific T cell reactivity. To assess the changes over time, blood and tissue are obtained during therapy to monitor therapy-induced changes in the periphery and the tumor microenvironment. Conclusion: Here we present our established workflow to systematically analyze tumor-host interactions in cancer patients and show preliminary data of different facets and clinical implications of these complex interactions. To evaluate and validate immunological and pathological characteristics of a tumor and the methods to assess its immunogenicity as well as its complex interplay with the microenvironment, comprehensive translational studies have to be conducted. Eventually, the obtained information will help to identify important subgroups and patients who are most likely to benefit from a certain type of immunotherapy. Citation Format: Zeynep Kosaloglu, Inka Zoernig, Niels Halama, Bruno Koehler, Anne Katrin Berger, Christoph Springfeld, Alexis Ulrich, Martin Schneider, Felix Lasitschka, Hanno Glimm, Stefan Froehling, Benedikt Brors, Dirk Jaeger. Combining immunomics and genomics for immunotherapy of refractory and rare cancers. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr LB-287.

The Nuanced Interplay of Intrinsic Disorder and Other Structural Properties Driving Protein Evolution.

Protein evolution often occurs at unequal rates in different sites along an amino acid chain. Site-specific evolutionary rates have been linked to several structural and functional properties of proteins. Previous analyses of this phenomenon have involved relatively small datasets and, in some cases, the interaction among multiple structural factors is not evaluated. Here, we present the results of a large-scale phylogenetic and statistical analysis, testing the effects and interactions of three structural properties on amino acid replacement rates. We used sequence-based computational methods to predict (i) intrinsic disorder propensity, (ii) secondary structure, and (iii) functional domain involvement across millions of amino acid sites in thousands of sequence alignments of metazoan proteins. Our results somewhat corroborate earlier findings that intrinsically disordered sites tend to be more variable than ordered sites, but there is considerable overlap among their rate distributions, and a significant confounding interaction exists between intrinsic disorder and secondary structure. Notably, protein sites that are consistently predicted to be both intrinsically disordered and involved in secondary structures tend to be the most conserved at the amino acid level, suggesting that they are highly constrained and functionally important. In addition, a significant interaction exists between functional domain involvement and secondary structure. These findings suggest that multiple structural drivers of protein evolution should be evaluated simultaneously in order to get a clear picture of their individual effects as well as any confounding interactions among them.

ADPRtool: A novel predicting model for identification of ASP-ADP-Ribosylation sites of human proteins.

Post-translational modifications (PTMs) occur in the vast majority of proteins, and they are essential for many protein functions. Computational prediction of the residue location of PTMs enhances the functional characterization of proteins. ADP-Ribosylation is an important type of PTM, because it is implicated in apoptosis, DNA repair, regulation of cell proliferation, and protein synthesis. However, mass spectrometric approaches have difficulties in identifying a vast number of protein ADP-Ribosylation sites. Therefore, a computational method for predicting ADP-Ribosylation sites of human proteins seems useful and necessary. Four types of sequence features and an incremental feature selection technique are utilized to predict protein ADP-Ribosylation sites. The final feature set for ADPR prediction modeling is optimized, based on a minimum redundancy maximum relevance criterion, so as to make more accurate predictions on aspartic acid ADPR modified residues. Our prediction model, ADPRtool, is capable to predict Asp-ADP-Ribosylation sites with a total accuracy of 85.45%, which is as good as most computational PTM site predictors. By using a sequence-based computational method, a new ADP-Ribosylation site prediction model - ADPRtool, is developed, and it has shown great accuracies with total accuracy, Matthew's correlation coefficient and area under receiver operating characteristic curve.

A method to predict the impact of regulatory variants from DNA sequence.

Most variants implicated in common human disease by Genome-Wide Association Studies (GWAS) lie in non-coding sequence intervals. Despite the suggestion that regulatory element disruption represents a common theme, identifying causal risk variants within indicted genomic regions remains a significant challenge. Here we present a novel sequence-based computational method to predict the effect of regulatory variation, using a classifier (gkm-SVM) which encodes cell-specific regulatory sequence vocabularies. The induced change in the gkm-SVM score, deltaSVM, quantifies the effect of variants. We show that deltaSVM accurately predicts the impact of SNPs on DNase I sensitivity in their native genomic context, and accurately predicts the results of dense mutagenesis of several enhancers in reporter assays. Previously validated GWAS SNPs yield large deltaSVM scores, and we predict novel risk SNPs for several autoimmune diseases. Thus, deltaSVM provides a powerful computational approach for systematically identifying functional regulatory variants.