SIFTER-T: a scalable and optimized framework for the SIFTER phylogenomic method of probabilistic protein domain annotation.

Statistical Inference of Function Through Evolutionary Relationships (SIFTER) is a powerful computational platform for probabilistic protein domain annotation. Nevertheless, SIFTER is not widely used, likely due to usability and scalability issues. Here we present SIFTER-T (SIFTER Throughput-optimized), a substantial improvement over SIFTER's original proof-of-principle implementation. SIFTER-T is optimized for better performance, allowing it to be used at the genome-wide scale. Compared to SIFTER 2.0, SIFTER-T achieved an 87-fold performance improvement using published test data sets for the known annotations recovering module and a 72.3% speed increase for the gene tree generation module in quad-core machines, as well as a major decrease in memory usage during the realignment phase. Memory optimization allowed an expanded set of proteins to be handled by SIFTER's probabilistic method. The improvement in performance and automation that we achieved allowed us to build a web server to bring the power of Bayesian phylogenomic inference to the genomics community. SIFTER-T and its online interface are freely available under GNU license at https://labpib.fmrp.usp.br/methods/SIFTER-t/ and https://github.com/dcasbioinfo/SIFTER-t.