DLPAlign

DLPAlign applies convolutional neural networks (CNNs) and bi-directional long short-term memory networks (Bi-LSTMs) to improve the accuracy of progressive multiple protein sequence alignment for evolutionary and structural analyses.

Key Features:

• Deep Learning Framework: Employs convolutional neural networks (CNNs) and bi-directional long short-term memory networks (Bi-LSTMs) to capture spatial hierarchies and sequential dependencies in protein sequences.
• Progressive Alignment Strategy: Implements a progressive alignment approach that iteratively aligns input sequences and calculates posterior probability matrices at each step.
• Empirical Performance Evaluation: Benchmarked against eleven leading multiple sequence alignment methods using BAliBASE, OXBench, and SABMark and reported superior total-column scores.
• Performance in Low Similarity Contexts: Improved alignment accuracy by approximately 2.8% over the second-best MSA software for families with average pairwise identity (PID) ≤30%.
• Application to Real-World Problems: Applied to protein secondary structure prediction for SARS-CoV-2 and outperformed other alignment tools across four related protein sequences.

Scientific Applications:

• Evolutionary Biology: Provides precise multiple sequence alignments for phylogenetic inference and comparative sequence analyses.
• Structural Bioinformatics: Supports structural modeling and protein secondary structure prediction through improved alignments.
• Functional Genomics: Enables alignment of divergent proteins with limited homology to assist functional annotation.
• Virology and Pandemic Research: Aids analysis of SARS-CoV-2 protein sequences for structure and function studies.

Methodology:

Builds a decision-making model using CNNs and Bi-LSTMs, performs progressive iterative alignment with calculation of posterior probability matrices at each step, and evaluates performance against BAliBASE, OXBench, and SABMark using total-column scores.