Home

Does in-distribution accuracy predict out-of-distribution accuracy for molecular models?
ALineMol helps you find out — with rigorous, reproducible distribution-shift evaluation for molecular property and activity prediction.

Get started Quickstart GitHub

---

Why ALineMol?

Molecular ML models are usually reported by their in-distribution (ID) accuracy — performance on a random test split. But in real drug discovery you deploy them on novel chemistry: new scaffolds, larger molecules, unexplored regions of chemical space. ALineMol makes it easy to measure how well ID performance transfers to these out-of-distribution (OOD) settings, and to study the agreement-on-the-line and accuracy-on-the-line phenomena for molecules.

•   16 splitting strategies

  ---

  Structure-, property-, clustering-, and similarity-based splitters behind one SMILES-first API. Generate realistic distribution shift with a single line.

  Splitting strategies

•   OOD evaluation

  ---

  Benchmark classical ML and graph neural networks on ID vs OOD data and quantify the generalization gap across datasets and models.

  OOD evaluation

•   Split quality analysis

  ---

  SplitAnalyzer measures train↔test similarity, scaffold overlap, and property divergence — so you can prove an "OOD" split is actually OOD.

  Split analysis

•   Runnable tutorials

  ---

  Colab-ready notebooks that take you from raw SMILES to a full ID/OOD comparison in minutes.

  Tutorials

A quick taste

from alinemol.splitters import get_splitter, SplitAnalyzer

smiles = ["CCO", "c1ccccc1", "CCN", "CC(=O)O", "c1ccncc1"]  # your dataset

# 1. Create an OOD split (Bemis-Murcko scaffold split)
splitter = get_splitter("scaffold", n_splits=5, test_size=0.2)
train_idx, test_idx = next(splitter.split(smiles))

# 2. Verify it is actually a distribution shift
analyzer = SplitAnalyzer(smiles)
report = analyzer.analyze_split(train_idx, test_idx, splitter_name="scaffold")
print(f"Mean train-test similarity: {report.similarity_metrics.mean_sim:.3f}")

Citation

If ALineMol is useful in your research, please cite the accompanying paper:

@article{fooladi2025alinemol,
  title   = {Evaluating Machine Learning Models for Molecular Property
             Prediction on Out-of-Distribution Data},
  author  = {Fooladi, Hosein and colleagues},
  journal = {Journal of Chemical Information and Modeling},
  year    = {2025},
  doi     = {10.1021/acs.jcim.5c00475}
}