Extracting phylogenetic dimensions of coevolution reveals hidden functional signals

Colavin, Alexandre; Atolia, Esha; Bitbol, Anne-Florence; Huang, Kerwyn Casey

doi:10.1038/s41598-021-04260-1

“…Separating coevolutionary signals encoding functional and structural constraints from phylogenetic correlations arising from historical contingency constitutes a key problem in analyzing the sequenceto-function mapping in proteins [15,18]. Phylogenetic correlations are known to obscure the identification of structural contacts by traditional coevolution methods, in particular by inferred Potts models [20,21,[41][42][43][44], motivating various corrections [17,21,22,24,[45][46][47][48]. From a theoretical point of view, disentangling these two types of signals is a fundamentally hard problem [49].…”

Section: Discussionmentioning

confidence: 99%

Protein language models trained on multiple sequence alignments learn phylogenetic relationships

Lupo

¹

,

Sgarbossa

²

,

Bitbol

³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Self-supervised neural language models with attention have recently been applied to biological sequence data, advancing structure, function and mutational effect prediction. Some protein language models, including MSA Transformer and AlphaFold's EvoFormer, take multiple sequence alignments (MSAs) of evolutionarily related proteins as inputs. Simple combinations of MSA Transformer's row attentions have led to state-of-the-art unsupervised structural contact prediction. We demonstrate that similarly simple, and universal, combinations of MSA Transformer's column attentions strongly correlate with Hamming distances between sequences in MSAs. Therefore, MSA-based language models encode detailed phylogenetic relationships. This could aid them to separate coevolutionary signals encoding functional and structural constraints from phylogenetic correlations arising from historical contingency. To test this hypothesis, we generate synthetic MSAs, either without or with phylogeny, from Potts models trained on natural MSAs. We demonstrate that unsupervised contact prediction is indeed substantially more resilient to phylogenetic noise when using MSA Transformer versus inferred Potts models.

show abstract

“…Thus, disentangling signals was much harder than in the minimal model, as the couplings from phylogeny make the model richer even in the absence of phylogeny in the data generation step. While this is a difficult problem, it could be partially addressed by applying phylogeny corrections to the inferred couplings [ 51 , 57 ]. This could also shed light on whether some of the useful signal from non-contact pairs is coming from collective functional constraints, similar to sectors in single proteins [ 29 , 57 , 85 ], an interesting possibility that was not explored here.…”

Section: Discussionmentioning

confidence: 99%

“…While this is a difficult problem, it could be partially addressed by applying phylogeny corrections to the inferred couplings [ 51 , 57 ]. This could also shed light on whether some of the useful signal from non-contact pairs is coming from collective functional constraints, similar to sectors in single proteins [ 29 , 57 , 85 ], an interesting possibility that was not explored here. Investigating the impact of imperfections in protein sequence alignment on partner inference would also be highly relevant, as well as including the possibility of one-to-many pairings and crosstalk.…”

Section: Discussionmentioning

confidence: 99%

“…Therefore, correlations from contacts and from phylogeny are both useful to predict protein-protein interaction partners among paralogs. This stands in contrast with the identification of structural contacts by DCA [ 5 , 6 , 30 , 49 – 51 ], where phylogenetic correlations obscure structural ones, motivating the use of phylogeny corrections [ 52 , 53 ], such as the Average Product Correction [ 54 , 55 ], reweighting close sequences [ 6 , 7 , 55 , 56 ], and Nested Coevolution [ 57 ].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Correlations from structure and phylogeny combine constructively in the inference of protein partners from sequences

Gerardos

¹

,

Dietler²,

Bitbol³

2022

Self Cite

View full text Add to dashboard Cite

Inferring protein-protein interactions from sequences is an important task in computational biology. Recent methods based on Direct Coupling Analysis (DCA) or Mutual Information (MI) allow to find interaction partners among paralogs of two protein families. Does successful inference mainly rely on correlations from structural contacts or from phylogeny, or both? Do these two types of signal combine constructively or hinder each other? To address these questions, we generate and analyze synthetic data produced using a minimal model that allows us to control the amounts of structural constraints and phylogeny. We show that correlations from these two sources combine constructively to increase the performance of partner inference by DCA or MI. Furthermore, signal from phylogeny can rescue partner inference when signal from contacts becomes less informative, including in the realistic case where inter-protein contacts are restricted to a small subset of sites. We also demonstrate that DCA-inferred couplings between non-contact pairs of sites improve partner inference in the presence of strong phylogeny, while deteriorating it otherwise. Moreover, restricting to non-contact pairs of sites preserves inference performance in the presence of strong phylogeny. In a natural data set, as well as in realistic synthetic data based on it, we find that non-contact pairs of sites contribute positively to partner inference performance, and that restricting to them preserves performance, evidencing an important role of phylogeny.

show abstract

“…Therefore, correlations from contacts and from phylogeny are both useful to predict protein-protein interaction partners among paralogs. This stands in contrast with the identification of structural contacts by DCA [5,6,30,[49][50][51], where phylogenetic correlations obscure structural ones, motivating the use of phylogeny corrections [52,53], such as the Average Product Correction [54,55], reweighting close sequences [6,7,55,56], and Nested Coevolution [57].…”

Section: Introductionmentioning

confidence: 99%

Correlations from structure and phylogeny combine constructively in the inference of protein partners from sequences

Gerardos

¹

,

Dietler

²

,

Bitbol

³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Inferring protein-protein interactions from sequences is an important task in computational biology. Recent methods based on Direct Coupling Analysis (DCA) or Mutual Information (MI) allow to find interaction partners among paralogs of two protein families. Does successful inference mainly rely on correlations from structural contacts or from phylogeny, or both? Do these two types of signal combine constructively or hinder each other? To address these questions, we generate and analyze synthetic data produced using a minimal model that allows us to control the amounts of structural constraints and phylogeny. We show that correlations from these two sources combine constructively to increase the performance of partner inference by DCA or MI. Furthermore, signal from phylogeny can rescue partner inference when signal from contacts becomes less informative, including in the realistic case where inter-protein contacts are restricted to a small subset of sites. We also demonstrate that DCA-inferred couplings between non-contact pairs of sites improve partner inference in the presence of strong phylogeny, while deteriorating it otherwise. Moreover, restricting to non-contact pairs of sites preserves inference performance in the presence of strong phylogeny. In a natural dataset, as well as in realistic synthetic data based on it, we find that non-contact pairs of sites contribute positively to partner inference performance, and that restricting to them preserves performance, evidencing an important role of phylogeny.Author summaryIn protein sequence data, the amino acid usages at different sites of a protein or of two interacting proteins can be correlated because of functional constraints. For instance, the need to maintain physico-chemical complementarity among two sites that are in contact in the three-dimensional structure of a protein complex causes such correlations. However, correlations can also arise due to shared evolutionary history, even in the absence of any functional constraint. While these phylogenetic correlations are known to obscure the inference of structural contacts, we show, using controlled synthetic data, that correlations from structure and phylogeny combine constructively to allow the inference of protein partners from sequences. We also show that pairs of amino acids that are not in contact in the structure have a major impact on partner inference in a natural dataset and in realistic synthetic ones. These findings explain the success of methods based on pairwise maximum-entropy models or on information theory at predicting protein partners from sequences.

show abstract

Extracting phylogenetic dimensions of coevolution reveals hidden functional signals

Cited by 15 publications

References 77 publications

Protein language models trained on multiple sequence alignments learn phylogenetic relationships

Protein language models trained on multiple sequence alignments learn phylogenetic relationships

Correlations from structure and phylogeny combine constructively in the inference of protein partners from sequences

Correlations from structure and phylogeny combine constructively in the inference of protein partners from sequences

Contact Info

Product

Resources

About