EvoLSTM: context-dependent models of sequence evolution using a sequence-to-sequence LSTM.

Area of Science:

• Bioinformatics and computational biology
• Genomics and evolutionary biology
• Machine learning applications

Background:

• Accurate probabilistic models of sequence evolution are crucial for bioinformatics tasks like sequence alignment and phylogenetic inference.
• Realistic simulation of sequence evolution is vital for benchmarking methods.
• Current models often fail to capture complex context dependencies in mutational processes.

Purpose of the Study:

• To introduce EvoLSTM, a novel recurrent neural network-based simulator for sequence evolution.
• To develop a tool that captures context dependencies in mutational processes.
• To enhance the realism of sequence evolution simulations.

Main Methods:

• Developed EvoLSTM, a sequence-to-sequence long short-term memory model.
• Trained the model to predict mutation probabilities considering 14 flanking nucleotides.
• Applied the model to simulate mammalian and plant DNA sequence evolution.

Main Results:

• EvoLSTM realistically simulates DNA sequence evolution in mammals and plants.
• The model captures complex mutational context dependencies.
• Unexpectedly strong long-range dependencies in mutation probabilities were revealed.

Conclusions:

• EvoLSTM effectively models context-dependent mutational processes using machine learning.
• The tool provides a more realistic approach to simulating sequence evolution.
• EvoLSTM is a valuable resource for studying complex evolutionary dynamics and improving bioinformatics tools.