Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Insights into the Unknown: Federated Data Diversity Analysis on Molecular Data

Markus Bujotzek, Evelyn Trautmann, Calum Hand, Ian Hales

TL;DR

This work tackles the challenge of assessing data diversity in privacy-preserving federated learning for drug discovery by benchmarking three federated clustering methods—Fed-kMeans, Fed-PCA+Fed-kMeans, and Fed-LSH—across eight molecular datasets and introducing SF-ICF, a chemistry-informed metric that links cluster structure to molecular scaffolds. The authors combine standard mathematical metrics with domain-specific evaluation and on-client explainability to show that while Fed-LSH excels on Euclidean clustering metrics, Fed-kMeans and Fed-PCA+Fed-kMeans align better with scaffold-based diversity via SF-ICF; federated approaches can approach, and occasionally exceed, centralized baselines. They also demonstrate the importance of explainability, revealing scaffold structures as the most informative feature group and highlighting the need for domain-aware metrics to interpret clustering in molecular space. Overall, the paper provides a foundation for trustworthy federated diversity analysis in AI-driven drug discovery, emphasizing chemistry-informed evaluation and local interpretability to guide deployment across data silos.

Abstract

AI methods are increasingly shaping pharmaceutical drug discovery. However, their translation to industrial applications remains limited due to their reliance on public datasets, lacking scale and diversity of proprietary pharmaceutical data. Federated learning (FL) offers a promising approach to integrate private data into privacy-preserving, collaborative model training across data silos. This federated data access complicates important data-centric tasks such as estimating dataset diversity, performing informed data splits, and understanding the structure of the combined chemical space. To address this gap, we investigate how well federated clustering methods can disentangle and represent distributed molecular data. We benchmark three approaches, Federated kMeans (Fed-kMeans), Federated Principal Component Analysis combined with Fed-kMeans (Fed-PCA+Fed-kMeans), and Federated Locality-Sensitive Hashing (Fed-LSH), against their centralized counterparts on eight diverse molecular datasets. Our evaluation utilizes both, standard mathematical and a chemistry-informed evaluation metrics, SF-ICF, that we introduce in this work. The large-scale benchmarking combined with an in-depth explainability analysis shows the importance of incorporating domain knowledge through chemistry-informed metrics, and on-client explainability analyses for federated diversity analysis on molecular data.

Insights into the Unknown: Federated Data Diversity Analysis on Molecular Data

TL;DR

This work tackles the challenge of assessing data diversity in privacy-preserving federated learning for drug discovery by benchmarking three federated clustering methods—Fed-kMeans, Fed-PCA+Fed-kMeans, and Fed-LSH—across eight molecular datasets and introducing SF-ICF, a chemistry-informed metric that links cluster structure to molecular scaffolds. The authors combine standard mathematical metrics with domain-specific evaluation and on-client explainability to show that while Fed-LSH excels on Euclidean clustering metrics, Fed-kMeans and Fed-PCA+Fed-kMeans align better with scaffold-based diversity via SF-ICF; federated approaches can approach, and occasionally exceed, centralized baselines. They also demonstrate the importance of explainability, revealing scaffold structures as the most informative feature group and highlighting the need for domain-aware metrics to interpret clustering in molecular space. Overall, the paper provides a foundation for trustworthy federated diversity analysis in AI-driven drug discovery, emphasizing chemistry-informed evaluation and local interpretability to guide deployment across data silos.

Abstract

AI methods are increasingly shaping pharmaceutical drug discovery. However, their translation to industrial applications remains limited due to their reliance on public datasets, lacking scale and diversity of proprietary pharmaceutical data. Federated learning (FL) offers a promising approach to integrate private data into privacy-preserving, collaborative model training across data silos. This federated data access complicates important data-centric tasks such as estimating dataset diversity, performing informed data splits, and understanding the structure of the combined chemical space. To address this gap, we investigate how well federated clustering methods can disentangle and represent distributed molecular data. We benchmark three approaches, Federated kMeans (Fed-kMeans), Federated Principal Component Analysis combined with Fed-kMeans (Fed-PCA+Fed-kMeans), and Federated Locality-Sensitive Hashing (Fed-LSH), against their centralized counterparts on eight diverse molecular datasets. Our evaluation utilizes both, standard mathematical and a chemistry-informed evaluation metrics, SF-ICF, that we introduce in this work. The large-scale benchmarking combined with an in-depth explainability analysis shows the importance of incorporating domain knowledge through chemistry-informed metrics, and on-client explainability analyses for federated diversity analysis on molecular data.
Paper Structure (17 sections, 4 equations, 4 figures, 3 tables)

This paper contains 17 sections, 4 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Data
  4. Federated Clustering Methods
  5. Federated k-Means Clustering
  6. Federated Principal Component Analysis
  7. Federated Locality-sensitive Hashing
  8. Evaluation metrics
  9. Experiments
  10. Results and Discussion
  11. Hyperparameter Grid Search
  12. Benchmarking Results
  13. Explainability
  14. Feature Group Importance
  15. Cluster Counts and Sizes
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Benchmarking results of federated clustering (Fed-kMeans, Fed-PCA+Fed-kMeans, Fed-LSH) methods against their centralized counterparts. The methods are compared w.r.t. 3 mathematical (Silhouette (Euclidean), Davies-Bouldin, Calinski-Harabasz), and the chemistry-informed evaluation metric (SF-ICF). Mean values are marked with a diamond, arrows point towards better metric values.
  • Figure 2: Feature group importances of meta information for federated clustering methods of FL client 1 of 5 on ames data.
  • Figure 3: Qualitative clustering results of federated clustering methods on FL client 1 of 5 on ames data. The high-dimensional data was dimension-reduced by a PCA with $p=3$.
  • Figure 4: Per-cluster SF–ICF versus scaffold-based KL divergence over all federated experiments. Each point represents a cluster; color encodes cluster size.