In silico learning of tumor evolution through mutational time series

Area of Science:

• Computational Biology
• Genomics
• Machine Learning

Background:

• Cancer develops from accumulated somatic mutations.
• Understanding mutation order aids early diagnosis and treatment decisions.

Purpose of the Study:

• To apply long short-term memory (LSTM) networks to model tumor mutational evolution.
• To predict mutation burden and sequence occurrence during cancer progression.

Main Methods:

• Utilized LSTM networks, a type of recurrent neural network, to analyze mutational time series.
• Simulated mutational data based on learned probabilities for statistical comparison with empirical data.
• Identified passenger mutations associated with cancer drivers and analyzed gene interaction networks.

Main Results:

• LSTMs accurately learned complex dynamics of tumor progression and predicted mutational burden.
• Simulated data was statistically indistinguishable from real-world mutational data.
• Discovered significant enrichment of interactions between passenger mutations and driver genes, linked to poor prognosis.

Conclusions:

• LSTM networks offer a powerful tool for predicting cancer evolution and identifying novel therapeutic targets.
• The study reveals previously unknown aspects of cancer development through driver-passenger gene interactions.