The field of protein function prediction as viewed by different domain scientists

Abstract

Motivation Experimental biologists, biocurators, and computational biologists all play a role in characterizing a protein’s function. The discovery of protein function in the laboratory by experimental scientists is the foundation of our knowledge about proteins. Experimental findings are compiled in knowledgebases by biocurators to provide standardized, readily accessible, and computationally amenable information. Computational biologists train their methods using these data to predict protein function and guide subsequent experiments. To understand the state of affairs in this ecosystem, centered here around protein function prediction, we surveyed scientists from these three constituent communities. Results We show that the three communities have common but also idiosyncratic perspectives on the field. Most strikingly, experimentalists rarely use state-of-the-art prediction software, but when presented with predictions, report many to be surprising and useful. Ontologies appear to be highly valued by biocurators, less so by experimentalists and computational biologists, yet controlled vocabularies bridge the communities and simplify the prediction task. Additionally, many software tools are not readily accessible and the predictions presented to the users can be broad and uninformative. We conclude that to meet both the social and technical challenges in the field, a more productive and meaningful interaction between members of the core communities is necessary. Supplementary information Supplementary data are available at Bioinformatics Advances online.

Identification of type VI secretion system effector-immunity pairs using structural bioinformatics

“…T6EC3 shares a catalytic glutamate residue with its homolog sleB that is essential for its toxicity. This proof-of-concept shows the utility of these novel structural bioinformatic tools, especially to experimental scientists (Ramola et al, 2022 ), for the discovery of novel T6E-immunity pairs.…”

COLLAPSE: A representation learning framework for identification and characterization of protein structural sites

“…Recently, methods have applied machine learning to predict function from sequence (Kulmanov & Hoehndorf, 2020; Sanderson et al, 2021) or structure (Gligorijević et al, 2021). However, like profile‐based methods, these lack the local resolution necessary to identify specific functional sites, and their reliance on nonspecific functional labels such as those provided by Gene Ontology terms (Ashburner et al, 2000) often limits practical utility (Ramola et al, 2022). Machine learning approaches that focus on local functional sites are either specific to a particular type of site (e.g., ligand binding, [Tubiana et al, 2022]; [Zhao et al, 2020] enzyme active sites [Moraes et al, 2017]) or require building specific models for each functional site of interest, (Buturovic et al, 2014; Torng & Altman, 2019a) which can be computationally expensive and demands sufficient data to train an accurate model.…”

Explainable protein function annotation using local structure embeddings

“…Moreover, without reliable identification of functional residues the global predictions themselves may be misleading; for example, consider an enzymatic domain which is lacking a single catalytic residue critical to its function and is therefore inactive. Indeed, the lack of methods which can make accurate global predictions and provide residue-level explainability is cited as a major reason why few newly developed functional predictors are widely adopted by experimental biologists (28).…”

“…First, to assemble sufficiently large labeled training datasets, many methods rely on pre-defined labels which are often broad or ambiguous. For example, GO terms have varying levels of specificity and single terms may overlap or contain multiple distinct biochemical functions (28). Similarly, although EC numbers are arranged in a four-level hierarchy with a more consistent level of specificity at the lowest level, some EC numbers are so rare that they are either excluded from training or aggregated up to a higher level of the tree.…”