Using Coevolutionary Scoring to Distinguish Among AlphaFold Models of the Pseudomonas aeruginosa Type III Secretion Translocon

Presenter: Arnon Kuzmin

Faculty Sponsor: Alejandro Heuck

School: UMass Amherst

Research Area: Biochemistry and Molecular Biology

Session: Poster Session 3, 1:15 PM - 2:00 PM, 165, D10

ABSTRACT

The Type III Secretion System translocon of Pseudomonas aeruginosa is essential for injecting effector proteins into host cells, but its hetero-oligomeric membrane architecture remains unknown. Determining its structure is essential for understanding how this virulence complex assembles and functions, making reliable computational models critical for prioritizing experimentally testable structural hypotheses. Standard AlphaFold-Multimer runs with default A3M inputs produced low-confidence and often inconsistent models. Analysis of AlphaFold’s A3M files revealed a very limited number of sequences, with numerous incorrectly paired or redundant sequences, motivating targeted curation of paired homologs. Incorporating curated paired sequences into default A3Ms produced models with improved AlphaFold confidence metrics. However, multiple top-scoring models often displayed distinct interface geometries, making it difficult to discriminate which structure is more plausible. To resolve such ties, we developed a coevolution-augmented scoring pipeline that computes residue-pair coupling scores from aligned homologs and re-ranks candidate models based on the number of coevolving residue pairs that are spatially proximate in each structure. Applied to ensembles of P. aeruginosa translocon models, it discriminated among top-scoring yet structurally distinct models. This coevolution-augmented ranking framework provides a practical, biologically informed method to select plausible models from AlphaFold-Multimer ensembles when conventional confidence metrics are ambiguous. We plan to benchmark the scoring pipeline on complexes with known structures, evaluate the impact of alternative curated MSAs, and advance top candidates for experimental validation.

RELATED ABSTRACTS