RESUMO
Thrombin is a key enzyme involved in the development and progression of many cardiovascular diseases. Direct thrombin inhibitors (DTIs), with their minimum off-target effects and immediacy of action, have greatly improved the treatment of these diseases. However, the risk of bleeding, pharmacokinetic issues, and thrombotic complications remain major concerns. In an effort to increase the effectiveness of the DTI discovery pipeline, we developed a two-stage machine learning pipeline to identify and rank peptide sequences based on their effective thrombin inhibitory potential. The positive dataset for our model consisted of thrombin inhibitor peptides and their binding affinities (KI) curated from published literature, and the negative dataset consisted of peptides with no known thrombin inhibitory or related activity. The first stage of the model identified thrombin inhibitory sequences with Matthew's Correlation Coefficient (MCC) of 83.6%. The second stage of the model, which covers an eight-order of magnitude range in KI values, predicted the binding affinity of new sequences with a log room mean square error (RMSE) of 1.114. These models also revealed physicochemical and structural characteristics that are hidden but unique to thrombin inhibitor peptides. Using the model, we classified more than 10 million peptides from diverse sources and identified unique short peptide sequences (<15 aa) of interest, based on their predicted KI. Based on the binding energies of the interaction of the peptide with thrombin, we identified a promising set of putative DTI candidates. The prediction pipeline is available on a web server.
RESUMO
A majority of microbial infections are associated with biofilms. Targeting biofilms is considered an effective strategy to limit microbial virulence while minimizing the development of antibiotic resistance. Toward this need, antibiofilm peptides are an attractive arsenal since they are bestowed with properties orthogonal to small molecule drugs. In this work, we developed machine learning models to identify the distinguishing characteristics of known antibiofilm peptides, and to mine peptide databases from diverse habitats to classify new peptides with potential antibiofilm activities. Additionally, we used the reported minimum inhibitory/eradication concentration (MBIC/MBEC) of the antibiofilm peptides to create a regression model on top of the classification model to predict the effectiveness of new antibiofilm peptides. We used a positive dataset containing 242 antibiofilm peptides, and a negative dataset which, unlike previous datasets, contains peptides that are likely to promote biofilm formation. Our model achieved a classification accuracy greater than 98% and harmonic mean of precision-recall (F1) and Matthews correlation coefficient (MCC) scores greater than 0.90; the regression model achieved an MCC score greater than 0.81. We utilized our classification-regression pipeline to evaluate 135,015 peptides from diverse sources for potential antibiofilm activity, and we identified 185 candidates that are likely to be effective against preformed biofilms at micromolar concentrations. Structural analysis of the top 37 hits revealed a larger distribution of helices and coils than sheets, and common functional motifs. Sequence alignment of these hits with known antibiofilm peptides revealed that, while some of the hits showed relatively high sequence similarity with known peptides, some others did not indicate the presence of antibiofilm activity in novel sources or sequences. Further, some of the hits had previously recognized therapeutic properties or host defense traits suggestive of drug repurposing applications. Taken together, this work demonstrates a new in silico approach to predicting antibiofilm efficacy, and identifies promising new candidates for biofilm eradication.