Using Machine Learning and Integrative Approaches for Current Problems in Structural Biology

Although the fundamental forces between atoms and molecules are almost fully understood at a theoretical level, and computer simulations have become an integral part of research activities, the application of these methods to large biomolecules still faces important practical difficulties due to the combinatorial explosion of possible interactions involved. Developing efficient protein structure prediction algorithms thus remains a major scientific challenge in computational biology.

In my talk will demonstrate how machine learning and optimization in general can be used to design interaction potentials adapted to specific conformational space exploration problems. More precisely, I will present our recent results on the prediction of properties of small molecules [1], on the prediction of protein-protein [2, 3, 4] and protein-drug interactions [5, 6], as well as individual protein folds at atomic level. I will also present our methods for efficient space exploration with the derived knowledge-based potentials including a novel FFT-accelerated technique for protein-protein interactions [2] and protein-ligand applications.

In the second part of my talk I will present integrative modeling applications developed in our group that use additional information coming from Cryo-EM and SAXS experiments, for example. More specifically, I will describe a very efficient Pepsi-SAXS method that computes SAXS profiles and is 5-50 times faster compared to other methods [7], its application to protein-protein docking experiments and SAXS-guided structure prediction. For example, using this approach we obtained some of the best quality predictions in the very recent CASP12 exercise for SAXS-assisted targets. I will also present our NOLB non-linear normal mode analysis and its GUI interface [8] that computes some lowest normal modes for a mid-size system at interactive rates. Finally, I will present our novel developments for exhaustive flexible fitting into Cryo-EM maps.

[1]  Maria Kadukova & Sergei Grudinin, Knodle: A Support Vector Machines-Based Automatic Perception of Organic Molecules from 3D Coordinates, Journal of Chemical Information and Modeling, 2016, 56 (8), pp.1410-1419. 

[2]  Emilie Neveu, David Ritchie, Petr Popov & Sergei Grudinin, PEPSI-Dock: a detailed data- driven protein-protein interaction potential accelerated by polar Fourier correlation, Bioinformatics, 2016, 32 (7), pp.i693-i701. 

[3]  Petr Popov & Sergei Grudinin, Knowledge of Native Protein-Protein Interfaces Is Sufficient To Construct Predictive Models for the Selection of Binding Candidates, Journal of Chemical Information and Modeling, 2015, 55 (10), pp.2242-2255. 

[4]  Marc F. Lensink et al., Prediction of homo- and hetero-protein complexes by protein docking and template-based modeling: a CASP-CAPRI experiment, Proteins, 2016, 84 (S1), pp.323-348. 

[5]  Sergei Grudinin, Maria Kadukova, Andreas Eisenbarth, Simon Marillet & Frederic Cazals, Predicting binding poses and affinities for protein-ligand complexes in the 2015 D3R Grand Challenge using a physical model with a statistical parameter estimation, Journal of Computer-Aided Molecular Design, 2016, 30 (9), pp.791-804. 

[6]  Sergei Grudinin, Petr Popov, Emilie Neveu & Georgy Cheremovskiy, Predicting Binding Poses and Affinities in the CSAR 2013-2014 Docking Exercises Using the Knowledge-Based Convex- PL Potential, Journal of Chemical Information and Modeling, 2016, 56 (6), pp.1053-1062. 

[7]  Sergei Grudinin et al. Pepsi-SAXS : an adaptive method for rapid and accurate computation of small-angle X-ray scattering profiles. Acta Cryst., 2017, D73, pp.449 – 464. https://team.inria.fr/nano-d/software/pepsi-saxs/
[8] Alexandre Hoffmann & Sergei Grudinin. NOLB: Nonlinear Rigid Block Normal Mode Analysis Method. JCTC, 2017, 13 (5), pp.2123-2134. https://team.inria.fr/nano-d/software/nolb-normal-modes/