Simulation and empirical evaluation of biologically-informed neural network performance

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning

Background:

• Biologically-informed neural networks (BiNNs) provide interpretable models for biological data.
• Understanding dataset requirements for optimal BiNN performance is crucial but lacking.
• Previous work utilized P-NET, a BiNN, for prostate cancer metastasis prediction using somatic data.

Purpose of the Study:

• To develop simulation frameworks for evaluating factors influencing BiNN performance.
• To assess the impact of integrating germline and somatic data on prostate cancer metastasis prediction.
• To provide a principled framework for benchmarking BiNNs and understanding their data dependencies.

Main Methods:

• Developed two simulation frameworks to test BiNN performance under varying conditions (signal type, strength, sparsity, sample size).
• Empirically evaluated the P-NET model by integrating germline and somatic data for prostate cancer metastasis prediction.
• Analyzed factors limiting BiNN performance, including sample size, signal strength, and feature sparsity.

Main Results:

• BiNN performance is limited by small sample size, weak signal strength, and extreme feature sparsity.
• Simulations suggest BiNNs may favor linear over nonlinear signals.
• P-NET showed poor performance on sparse germline data; integration with somatic data did not improve prediction but enhanced interpretation.

Conclusions:

• Simulation frameworks enable systematic evaluation of dataset characteristics affecting BiNNs.
• Germline data integration did not improve predictive accuracy but enhanced model interpretability.
• These findings provide a foundation for developing and benchmarking BiNNs for biological data analysis.