AliFilter: a machine learning approach to alignment filtering.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multiple sequence alignments are vital for predicting protein structure/function and inferring phylogenetic trees.
• Highly divergent alignments often contain noise, necessitating filtering to remove poorly aligned or uninformative columns.
• Manual alignment curation is accurate but labor-intensive and less reproducible, while automated tools may lack adaptability.

Purpose of the Study:

• To develop an automated alignment curation tool that bridges manual and automated filtering approaches.
• To leverage machine learning for accurate and reproducible alignment filtering.
• To provide a flexible tool adaptable to various datasets and user-defined filtering criteria.

Main Methods:

• AliFilter employs a supervised machine learning approach for alignment curation.
• It trains a model on manually annotated alignments to automate the filtering process.
• Users can utilize a default model or create customized models for specific datasets.

Main Results:

• AliFilter accurately reproduces manual annotations with 98% accuracy.
• The tool is resilient to errors in the training data.
• In phylogenomic workflows, AliFilter reduced runtime by 35% with minimal impact on alignment results.

Conclusions:

• AliFilter offers an accurate, reproducible, and efficient method for automated alignment curation.
• It successfully automates manual filtering tasks, saving significant computational time.
• AliFilter is free, open-source software available for multiple operating systems.