MLProbs: A Data-Centric Pipeline for Better Multiple Sequence Alignment.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Traditional Multiple Sequence Alignment (MSA) construction relies on algorithm-centric approaches, often reducing the problem to complex combinatorial optimization.
• These methods may not optimally handle the inherent variability and complexity of biological sequence data.
• A data-centric approach, leveraging machine learning on benchmark datasets, offers a promising alternative.

Purpose of the Study:

• To develop and evaluate a novel data-centric pipeline for Multiple Sequence Alignment (MSA) construction.
• To demonstrate the efficacy of shallow machine learning models in guiding MSA tool selection and realignment decisions.
• To improve MSA accuracy, particularly for challenging datasets like low-similarity protein families.

Main Methods:

• Developed MLProbs, a new MSA pipeline based on a data-centric approach.
• Trained shallow machine learning classification models on benchmark data to guide alignment tool choice and realignment.
• Evaluated MLProbs against 10 popular MSA tools using four benchmark databases (BAliBASE, OXBench, OXBench-X, SABMark).

Main Results:

• MLProbs consistently achieved the highest TC score across benchmark databases.
• Demonstrated significant improvement for protein families with low similarity (≤ 50%), outperforming top competitors by over 1.8%.
• MLProbs exhibited superior performance in real-life applications, including phylogenetic tree construction and protein secondary structure prediction.

Conclusions:

• The data-centric approach, powered by shallow machine learning, offers a robust and effective strategy for Multiple Sequence Alignment.
• MLProbs provides a significant advancement in MSA accuracy, especially for evolutionarily distant sequences.
• Future research could explore deep learning methods to further enhance MSA construction capabilities.