A huge thanks to my fantastic multi-disciplinary co-authors: Vladimir A. Shitov, Malte Lücken, Niki Kilbertus, Stefan Bauer, Marie Piraud, Alena Buyx, and Fabian Theis.

To fighting biases in ML-based single-cell science!

Our work traces these biases’ origins and interactions across the development pipeline. This pipeline-informed approach highlights how biases interconnect, potentially amplifying their impacts and complicating mitigation efforts.

Biases arise at every step of the ML-based single-cell analysis pipeline. We highlight:
🏥 Clinical Biases
👥 Cohort biases
🧪 Biases introduced during single-cell sequencing
🤖 Machine-learning and interpretational biases specific to weakly supervised or unsupervised ML models

A brief TL;DR:
Recent advances in ML-based single-cell data science offer groundbreaking insights into human health, enabling the stratification of tissue donors at single-cell resolution. But these insights are not immune to biases that can compromise their generalizability and fairness.

Biases arise at every step of the ML-based single-cell analysis pipeline. We highlight:
🏥 Clinical Biases
👥 Cohort biases
🧪 Biases introduced during single-cell sequencing
🤖 Machine-learning and interpretational biases specific to weakly supervised or unsupervised ML models

A brief TL;DR:
Recent advances in ML-based single-cell data science offer groundbreaking insights into human health, enabling the stratification of tissue donors at single-cell resolution. But these insights are not immune to biases that can compromise their generalizability and fairness.