Unsupervised and Supervised Learning over the Energy Landscape for Protein Decoy Selection

Abstract

The energy landscape that organizes microstates of a molecular system and governs the underlying molecular dynamics exposes the relationship between molecular form/structure, changes to form, and biological activity or function in the cell. However, several challenges stand in the way of leveraging energy landscapes for relating structure and structural dynamics to function. Energy landscapes are high-dimensional, multi-modal, and often overly-rugged. Deep wells or basins in them do not always correspond to stable structural states but are instead the result of inherent inaccuracies in semi-empirical molecular energy functions. Due to these challenges, energetics is typically ignored in computational approaches addressing long-standing central questions in computational biology, such as protein decoy selection. In the latter, the goal is to determine over a possibly large number of computationally-generated three-dimensional structures of a protein those structures that are biologically-active/native. In recent work, we have recast our attention on the protein energy landscape and its role in helping us to advance decoy selection. Here, we summarize some of our successes so far in this direction via unsupervised learning. More importantly, we further advance the argument that the energy landscape holds valuable information to aid and advance the state of protein decoy selection via novel machine learning methodologies that leverage supervised learning. Our focus in this article is on decoy selection for the purpose of a rigorous, quantitative evaluation of how leveraging protein energy landscapes advances an important problem in protein modeling. However, the ideas and concepts presented here are generally useful to make discoveries in studies aiming to relate molecular structure and structural dynamics to function.

Fewer Dimensions, More Structures for Improved Discrete Models of Dynamics of Free versus Antigen-Bound Antibody

“…Further details can be found in [ 54 ]. We note that basin finding over conformation-energy samples has been employed by several works for molecular structure–function studies [ 12 , 55 , 56 , 57 ].…”

Physicochemical Heuristics for Identifying High Fidelity, Near-Native Structural Models of Peptide/MHC Complexes

“…First, some methods that predict peptide affinity for MHC proteins consider position-specific features determined by the different structures and chemistries of the pockets that line the MHC binding groove ( 9 , 63 ). Second, efforts to improve protein structure prediction have explored using machine learning to selectively weight terms in energy functions, resulting in optimized decoy selections approaches trained for specific systems or tasks ( 28 , 33 , 34 , 64 – 66 ). Accordingly, we explored regression approaches in which peptide position-dependent structural and energetic terms were differentially weighted to yield functions optimized for identifying near-native decoys of nonamers bound to HLA-A*02:01.…”

Decoy Selection for Protein Structure Prediction Via Extreme Gradient Boosting and Ranking

“…Even for these challenging decoy sets, ML-Select significantly outperforms the four existing selection strategies in 5 out of 7 test cases (1hhp, 2ezk, 1aoy, 2h5nd, and 1aly) for all sizes of basin selections (i.e., B 1−x , x ∈ [1 − 3]). For two other cases (1isua and 1cc5 ), ML-Select performs better for the top basin for 1isua, and for 1cc5 when x ∈ [2,3]. For instance, for the most difficult test case 1aly, ML-Select obtains about 42% purity whereas the four other methods fail to provide a single true positive (0% purity).…”

“…The basins, extracted from the energy landscape, can be useful in decoy selection. Works in [3,4] shows that simple, ranking-based basin selection strategies outperform a standard clustering-based decoy selection method in terms of purity (percentage of true positives, penalizes the selected basin by the extent of false positives found in that basin). Basins can be ranked as a combination of basin characteristics.…”