EvoLSTM

EvoLSTM models context-dependent DNA sequence evolution by using a sequence-to-sequence Long Short-Term Memory (LSTM) recurrent neural network to predict position-specific mutation probabilities for evolutionary and comparative analyses.

Key Features:

• Context-Dependent Modeling: EvoLSTM captures context dependencies by considering each nucleotide plus its 14 flanking nucleotides to model mutational processes.
• Probabilistic Framework: It outputs position-specific mutation probability estimates at each sequence position for use in downstream analyses.
• Simulation Capabilities: EvoLSTM simulates DNA sequence evolution in mammalian and plant genomes and can reveal strong long-range context dependencies.
• Machine Learning Integration: It employs a sequence-to-sequence LSTM recurrent neural network architecture to learn mutational dynamics from sequence context.

Scientific Applications:

• Sequence Alignment and Phylogenetic Inference: Provides mutation probability predictions to refine alignment scoring and inform phylogenetic models.
• Benchmarking Strategies: Generates realistic simulated datasets for developing and testing benchmarking strategies in bioinformatics.
• Studying Mutational Processes: Enables analysis of complex and long-range context dependencies that influence mutation probabilities.

Methodology:

EvoLSTM is trained using a sequence-to-sequence LSTM model that learns to predict mutation probabilities by analyzing sequences and their surrounding nucleotide contexts.