Kalakoti, Y. | bioRxiv Science

bioRxiv2024

AFsample2: Predicting multiple conformations and ensembles with AlphaFold2

Understanding protein dynamics and conformational states carries profound scientific and practical implications for several areas of research, ranging from a general understanding of biological processes at the molecular level to a detailed understanding of disease mechanisms, which in turn can open up new avenues in drug development. Multiple solutions have been recently developed to widen the conformational landscape of predictions made by Alphafold2 (AF2). Here, we introduce AFsample2, a method employing random MSA column masking to reduce the influence of co-evolutionary signals to enhance the structural diversity of models generated by the AF2 neural network. AFsample2 improves the prediction of alternative states for a broad range of proteins, yielding high-quality end states and diverse conformational ensembles. In the data set of open-closed conformations (OC23), alternate state models improved in 17 out of 23 cases without compromising the generation of the preferred state. Consistent results were observed in 16 membrane protein transporters, with improvements in 12 out of 16 targets. TM-score improvements to experimental end states were substantial, sometimes exceeding 50%, elevating mediocre scores from 0.58 to nearly perfect 0.98. Furthermore, AFsample2 increased the diversity of intermediate conformations by 70% compared to the standard AF2 system, producing highly confident models, that could potentially be on-path between the two states. In addition, we also propose a way of selecting the end-states in generated model ensembles. These solutions could potentially enhance the generation and identification of alternative protein conformations, thereby providing a more comprehensive understanding of protein function and dynamics. Future work will focus on validating the accuracy of these intermediate conformations and exploring their relevance to functional transitions in proteins.

bioinformatics↗

bioRxiv2023

Estimating protein-ligand interactions with geometric deep learning and mixture density models

Understanding the interactions between a ligand and its molecular target is crucial in guiding the optimization of molecules for any in-silico drug-design workflow. Multiple experimental and computational methods have been developed to better understand these intermolecular interactions. With the availability of a large number of structural datasets, there is a need for developing statistical frameworks that improve upon existing physics-based solutions. Here, we report a method based on geometric deep learning that is capable of predicting the binding conformations of ligands to protein targets. A technique to generate graphical representations of protein was developed to exploit the topological and electrostatic properties of the binding region. The developed framework, based on graph neural networks, learns a statistical potential based on the distance likelihood, which is tailor-made for each ligand-target pair. This potential can be coupled with global optimization algorithms such as differential evolution to reproduce the experimental binding conformations of ligands. We show that the potential based on distance likelihood, described here, performs similarly or better than well-established scoring functions for docking and screening tasks. Overall, this method represents an example of how artificial intelligence can be used to improve structure-based drug design. Significance statementCurrent machine learning-based solutions to model protein-ligand interactions lack the level of interpretability that physics-based methods usually provide. Here, a workflow to embed protein binding surfaces as graphs was developed to serve as a viable data structure to be processed by geometric deep learning. The developed architecture based on mixture density models was employed to accurately estimate the position and conformation of the small molecule within the binding region. The likelihood-based scoring function was compared against existing physics-based alternatives, and significant performance improvements in terms of docking power, screening power and reverse screening power were observed. Taken together, the developed framework provides a platform for utilising geometric deep-learning models for interpretable prediction of protein-ligand interactions at a residue level.

bioinformatics↗