Collective posterior inference from highly variable empirical replicates

Area of Science:

• Computational Biology
• Evolutionary Biology
• Statistical Inference

Background:

• High-throughput experiments generate large datasets from multiple observations.
• Standard simulation-based inference (SBI) methods face challenges scaling to noisy, multiple-replicate data due to computational costs and hyperparameter tuning.
• Existing SBI approaches struggle with outlier mitigation in complex datasets.

Purpose of the Study:

• To introduce a novel, fast, and robust method for collective posterior inference from multiple independent replicates.
• To address the limitations of current SBI methods in handling noisy and large-scale datasets.
• To provide a scalable and efficient solution for parameter estimation in evolutionary biology and other fields.

Main Methods:

• Developed a robust product-of-experts aggregation scheme for collective posterior inference.
• The new method automatically mitigates the influence of outliers in the data.
• Evaluated the method on both synthetic and empirical evolutionary datasets.

Main Results:

• Achieved state-of-the-art estimation accuracy and computational efficiency.
• Demonstrated successful inference even with noisy observations.
• The proposed method is compatible with existing SBI frameworks.

Conclusions:

• The new collective posterior inference method offers a scalable and plug-and-play solution for analyzing noisy multiple-replicate datasets.
• This approach enhances the robustness and efficiency of parameter estimation in complex biological systems.
• The method provides a significant advancement for simulation-based inference in high-throughput data analysis.