Cancer affects 20 million people worldwide; therefore, accelerated and urgent pathways are needed for the search for new drugs. The reactivation of cell death induced by the accumulation of iron-dependent lipid peroxides (Ferroptosis) is a potential mechanism for cancer treatment. One pathway to trigger Ferroptosis involves the inactivation of the glutathione-dependent peroxidase (GPx4) protein. Molecules with high affinity for iron are also being studied against cancer through chelation therapy. Therefore, the rapid detection of such molecules in molecular databases is of interest. Catechols exhibit a high iron-chelating power and can be synthesized by microorganisms as a chemical strategy for acquiring Fe3+. They can also be found in natural products. Recently, the natural product database in Peru (PeruNPDB) has been published where we found catechols; however, these have not yet been studied in GPx4. To expedite the search for catechols with affinity for the active site of GPx4, in large databases, we have developed a prediction model for scoring based on artificial neural networks. Our initial database consisted of the 83 catechols and 1,024 molecular descriptors. The target is the docking score obtained by ligand-receptor GPx4 coupling with the Autodock Vina software. Emphasizing the application of recursive feature elimination to improve the model’s performance and provides both the model and the procedure to be applied and improved by the scientific community. Our program enables the experimenter to rapidly filter molecules with high affinity specifically for the active site of GPx4, in the search for potential inhibitors.

Due to the growth computational power and easy access to vast amounts of biological data, machine learning (ML) models are being increasingly considered within the field of Rational Drug Design (RDD). In RDD, molecular docking methods are employed to predict the optimal binding mode of a small molecule (ligand) in the binding site of a target receptor (typically a protein). Molecular docking software utilizes Scoring Functions (SFs) to estimate the protein-ligand binding affinity. These SFs can be classified according to their development methodology: Physics-based, Empirical, Knowledge-based, and ML-based. ML-based SFs are trained using features extracted from protein-ligand complexes with known experimental affinities, relying heavily on the attributes used during training. Both the input data and the selected features can significantly impact the accuracy of the SFs. Additionally, different ML algorithms can be applied to select the most appropriate features and propose the SF model. In this study we propose RFL-Score, a novel ML-based SF using Random Forest with Lasso regression. RFL-Score is a ready-to-use SF developed with a straightforward methodology and trained exclusively on open-access biomolecular data. Only open-source software was employed for extracting features from the proteins, ligands, and complexes. The proposed SF was evaluated using the well-established CASF-2016 benchmark achieving outstanding performance compared to other SFs. The RFL-Score scoring function and elements used for training are freely available on GitHub at https://github.com/rflscore.

MOTIVATION: Despite the progress made in studying protein-ligand interactions and the widespread application of docking and affinity prediction tools, improving their precision and efficiency still remains a challenge. Computational approaches based on the scoring of docking conformations with statistical potentials constitute a popular alternative to more accurate but costly physics-based thermodynamic sampling methods. In this context, a minimalist and fast sidechain-free knowledge-based potential with a high docking and screening power can be very useful when screening a big number of putative docking conformations. RESULTS: Here, we present a novel coarse-grained potential defined by a 3D joint probability distribution function that only depends on the pairwise orientation and position between protein backbone and ligand atoms. Despite its extreme simplicity, our approach yields very competitive results with the state-of-the-art scoring functions, especially in docking and screening tasks. For example, we observed a twofold improvement in the median 5% enrichment factor on the DUD-E benchmark compared to Autodock Vina results. Moreover, our results prove that a coarse sidechain-free potential is sufficient for a very successful docking pose prediction. AVAILABILITYAND IMPLEMENTATION: The standalone version of KORP-PL with the corresponding tests and benchmarks are available at https://team.inria.fr/nano-d/korp-pl/ and https://chaconlab.org/modeling/korp-pl. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: One of the goals of the Systems Biology community is to have a detailed map of all biological interactions in an organism. One small yet important step in this direction is the creation of biological networks from post-genomic data. Bayesian networks are a very promising model for the inference of regulatory networks in Systems Biology. Usually, Bayesian networks are sampled with a Markov Chain Monte Carlo (MCMC) sampler in the structure space. Unfortunately, conventional MCMC sampling schemes are often slow in mixing and convergence. To improve MCMC convergence, an alternative method is proposed and tested with different sets of data. Moreover, the proposed method is compared with the traditional MCMC sampling scheme. RESULTS: In the proposed method, a simpler and faster method for the inference of regulatory networks, Graphical Gaussian Models (GGMs), is integrated into the Bayesian network inference, trough a Hierarchical Bayesian model. In this manner, information about the structure obtained from the data with GGMs is taken into account in the MCMC scheme, thus improving mixing and convergence. The proposed method is tested with three types of data, two from simulated models and one from real data. The results are compared with the results of the traditional MCMC sampling scheme in terms of network recovery accuracy and convergence. The results show that when compared with a traditional MCMC scheme, the proposed method presents improved convergence leading to better network reconstruction with less MCMC iterations. CONCLUSIONS: The proposed method is a viable alternative to improve mixing and convergence of traditional MCMC schemes. It allows the use of Bayesian networks with an MCMC sampler with less iterations. The proposed method has always converged earlier than the traditional MCMC scheme. We observe an improvement in accuracy of the recovered networks for the Gaussian simulated data, but this improvement is absent for both real data and data simulated from ODE.

BACKGROUND: This paper addresses the prediction of the free energy of binding of a drug candidate with enzyme InhA associated with Mycobacterium tuberculosis. This problem is found within rational drug design, where interactions between drug candidates and target proteins are verified through molecular docking simulations. In this application, it is important not only to correctly predict the free energy of binding, but also to provide a comprehensible model that could be validated by a domain specialist. Decision-tree induction algorithms have been successfully used in drug-design related applications, specially considering that decision trees are simple to understand, interpret, and validate. There are several decision-tree induction algorithms available for general-use, but each one has a bias that makes it more suitable for a particular data distribution. In this article, we propose and investigate the automatic design of decision-tree induction algorithms tailored to particular drug-enzyme binding data sets. We investigate the performance of our new method for evaluating binding conformations of different drug candidates to InhA, and we analyze our findings with respect to decision tree accuracy, comprehensibility, and biological relevance. RESULTS: The empirical analysis indicates that our method is capable of automatically generating decision-tree induction algorithms that significantly outperform the traditional C4.5 algorithm with respect to both accuracy and comprehensibility. In addition, we provide the biological interpretation of the rules generated by our approach, reinforcing the importance of comprehensible predictive models in this particular bioinformatics application. CONCLUSIONS: We conclude that automatically designing a decision-tree algorithm tailored to molecular docking data is a promising alternative for the prediction of the free energy from the binding of a drug candidate with a flexible-receptor.