Benchmarking computational tools for de novo motif discovery

Abstract

STRUCTURED ABSTRACTMotivationHigh-throughput experimental methods aimed at uncovering domain-peptide interactions have created numerous use cases for bioinformatic tools capable of detecting short linear motifs (SLiMs). Over the past twenty years, countless motif discovery tools have been introduced. However, the evaluation of these tools generally focus on a single tool and use artificially generated test-sets that fail to capture the authentic context of motifs within proteomic data. Consequently, these evaluations fall short in addressing real-world use cases.ResultsIn the current work, five motif discovery tools and seven general sequence alignment tools were benchmarked on their capacity to find SLiMs in datasets of peptides of varying complexity built from curated SLiM instances from the Eukaryotic Linear Motif database. We explored the effect of dataset size, peptide lengths, background noise levels and motif complexity on the motif discovery of these tools. The main metric used to compare all tools outputs was the percentage of correctly aligned SLiM containing peptides. Two motif discovery tools (DEME and SLiMFinder) and a sequence alignment tool (OPAL) outperformed the rest of the tools when benchmarked with this metric, averaging over 70% correctly aligned motif-containing peptides. The performance of the motif discovery tools and OPAL were not affected by the sizes of the test-sets, while increasing peptide length and noise levels lowered the performances of all tools. For N-/C-terminal motifs all tools performed well, while for those motifs that had low motif complexity only DEME and SLiMFinder returned correctly aligned motifs for ∼50% or more of the datasets.Availabilityhttps://doi.org/10.5281/zenodo.10467208.