Efficient Bayesian inference for mechanistic modelling with high-throughput data

Area of Science:

• Computational Biology
• Systems Biology
• Biophysics

Background:

• Bayesian methods are crucial for integrating experimental data with mathematical models.
• High computational costs hinder Bayesian analysis of large datasets and complex models.
• Efficient computational strategies are needed for modern biological research.

Purpose of the Study:

• To introduce a computationally efficient minibatch approach for approximate Bayesian computation.
• To apply this method to analyze high-throughput imaging data from a scratch assay.
• To characterize gene knockdown effects on cell motility, proliferation, and wound healing.

Main Methods:

• Developed a minibatch approximation to Bayesian computation, inspired by Stochastic Gradient Descent.
• Applied a detailed mathematical model of cell dynamics (motility, proliferation, death).
• Analyzed data from a high-throughput imaging scratch assay involving 118 gene knockdowns.

Main Results:

• The minibatch approach significantly reduced computational cost compared to traditional Bayesian inference.
• Identified distinct functional subgroups of gene knockdowns based on cellular behavior.
• Characterized density-dependent and -independent motility and proliferation patterns.

Conclusions:

• Density-dependent interactions are critical for wound healing processes.
• The new computational method enables efficient analysis of complex biological systems.
• This approach facilitates the study of gene functions in cellular dynamics and tissue repair.