Embeddings from deep learning transfer GO annotations beyond homology

Maria Littmann,Burkhard Rost,Michael Heinzinger,Christian Dallago,Tobias Olenyi

doi:10.1038/s41598-020-80786-0

Maria Littmann, Burkhard Rost + Show 3 more

Open Access

https://doi.org/10.1038/s41598-020-80786-0

Copy DOI

Abstract

Knowing protein function is crucial to advance molecular and medical biology, yet experimental function annotations through the Gene Ontology (GO) exist for fewer than 0.5% of all known proteins. Computational methods bridge this sequence-annotation gap typically through homology-based annotation transfer by identifying sequence-similar proteins with known function or through prediction methods using evolutionary information. Here, we propose predicting GO terms through annotation transfer based on proximity of proteins in the SeqVec embedding rather than in sequence space. These embeddings originate from deep learned language models (LMs) for protein sequences (SeqVec) transferring the knowledge gained from predicting the next amino acid in 33 million protein sequences. Replicating the conditions of CAFA3, our method reaches an Fmax of 37 ± 2%, 50 ± 3%, and 57 ± 2% for BPO, MFO, and CCO, respectively. Numerically, this appears close to the top ten CAFA3 methods. When restricting the annotation transfer to proteins with < 20% pairwise sequence identity to the query, performance drops (Fmax BPO 33 ± 2%, MFO 43 ± 3%, CCO 53 ± 2%); this still outperforms naïve sequence-based transfer. Preliminary results from CAFA4 appear to confirm these findings. Overall, this new concept is likely to change the annotation of proteins, in particular for proteins from smaller families or proteins with intrinsically disordered regions.

Highlights

All organisms rely on the correct functioning of their cellular workhorses, namely their proteins involved in almost all roles, ranging from molecular functions (MF) such as chemical catalysis of enzymes to biological processes or pathways (BP), e.g., realized through signal transduction
In the sense that the database with annotations to transfer (GOA2017) had been available before the CAFA3 submission deadline (February 2017), our predictions were directly comparable to C AFA319
This embedding-based annotation transfer clearly outperformed the two CAFA3 baselines (Fig. 1: the simple BLAST for sequence-based annotation transfer, and the Naïve method assigning Gene Ontology (GO) terms statistically based on database frequencies, here GOA2017); it would have performed close to the top ten CAFA3 competitors had the method competed at CAFA3

Summary

Introduction

All organisms rely on the correct functioning of their cellular workhorses, namely their proteins involved in almost all roles, ranging from molecular functions (MF) such as chemical catalysis of enzymes to biological processes or pathways (BP), e.g., realized through signal transduction. In more formal terms: given a query Q of unknown and a template T of known function (Ft): IF PIDE(Q,T) > threshold θ, transfer annotation F t to Q. Instead of transferring annotations from the labeled protein T with the highest percentage pairwise sequence identity (PIDE) to the query Q, we chose T as the protein with the smallest distance in embedding space (DISTemb) to Q. This distance proxied the reliability of the prediction serving as threshold above which hits are considered too distant to infer annotations. The LMs were never trained on GO terms, we hypothesize LM embeddings to implicitly encode information relevant for the transfer of annotations, i.e., capturing aspects of protein function because embeddings have been shown to capture rudimentary features of protein structure and function[20,21,26,27]

Methods

Results

Conclusion