All the codes to reproduce the paper.

For now, please cite the preprint version.

• Milad Rayka, milad.rayka@yahoo.com

• Masoumeh Shams, masoumehshams.gh@gmail.com

1- Clone hca_ml Github repository.

git clone https://github.com/miladrayka/hca_ml.git

2- Change directory to hca_ml and make a new environment from the cheminf_env.yaml file by Mamba package manager:

mamba env create -f cheminf_env.yaml

To reproduce all results, tables, and figures, uncompress the Data.tar.xz and Results.tar.xz folders and refer to workflow.ipynb.

CAInsight is an interpretable and uncertainty-aware machine learning software designed to predict the activity of human carbonic anhydrase (hCA) isoforms. Specifically, we focus on predicting the activity of three isoforms: hCA II, hCA IX, and hCA XII.

The primary model relies on a Support Vector Machine (SVM) in conjunction with an Extended Connectivity Fingerprint (ECFP). Each hCA isoform has its own SVM-ECFP binary classifier that returns labels indicating whether they are active or inactive. We enhance our models with conformal prediction (CP), which quantifies the uncertainty in our predictions. In this context, CP can return an active label, an inactive label, a combination of both labels, or an empty set, depending on a specified epsilon value. Lastly, we employ counterfactual explainability (see exmol) to enhance the interpretability of our model.

To run CAInsight, change directory to hca_ml, then type the following in the terminal:

streamlit run gui.py

Copyright (c) 2025, Milad Rayka, Masoumeh Shams