TropMol-Caipora: A Cloud-Based Web Tool to Predict Cruzain Inhibitors by Machine Learning

Chagas disease (CD) affects approximately 8 million people and is classified as a high-priority neglected tropical disease by the WHO research and development actions. One promising avenue for drug development for CD is the inhibition of cruzain, a crucial cysteine protease of <i>T. cruzi</i> and one of the most extensively studied therapeutic targets. This study aims to construct a generic molecular screening model for public, online, and free use, based on pIC₅₀ cruzain predictions using a Random Forest model. For this, a data set with approximately 8 thousand compounds and 168 classes of descriptors was used, resulting in more than a million calculated descriptors. The model achieved <i>R</i> ² = 0.91 (RMSE = 0.33) for the training set and <i>R</i> ² = 0.72 (RMSE = 0.55) for the test set. In 5-fold cross-validation, performance remained consistent (<i>R</i> ² = 0.72 ± 0.01; RMSE = 0.57 ± 0.01). Some relevant insights were also observed. 1 - Aromaticity was shown to be a key factor in inhibitory activity. Compounds with nitrogenous aromatic rings are more likely to be more effective inhibitors. Aromatics in general also present correlation and structural relevance for an effective inhibitor. 2 - Halogenation may favor activity. The positive correlation may suggest that the introduction of halogen atoms may improve the activity of the compounds. 3 - Bicyclic or very rigid structures may decrease the inhibition efficiency of the tested candidates. 4 - Molecular accessibility and charge influence activity. Available in: https://colab.research.google.com/drive/1hotsXPddbJ6E0_hysLT9AqsXL-74Na-z?usp=sharing.