# Activity Cliff Prediction: Dataset and Benchmark

arXiv: https://arxiv.org/abs/2302.07541 · HTML: https://mlsanity.com/paper/2302.07541/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Builds on 15 of its 56 references that are in the map.
- First posted on arXiv 15 Feb 2023.

## Builds on (most cited first, 15 of 15)

- [Semi-Supervised Classification with Graph Convolutional Networks](https://mlsanity.com/paper/1609.02907/) (2016)
- [How Powerful are Graph Neural Networks?](https://mlsanity.com/paper/1810.00826/) (2018)
- [Simplifying Graph Convolutional Networks](https://mlsanity.com/paper/1902.07153/) (2019)
- [On the Properties of Neural Machine Translation: Encoder-Decoder Approaches](https://mlsanity.com/paper/1409.1259/) (2014)
- [MoleculeNet: A Benchmark for Molecular Machine Learning](https://mlsanity.com/paper/1703.00564/) (2017)
- [Strategies for Pre-training Graph Neural Networks](https://mlsanity.com/paper/1905.12265/) (2019)
- [Generalizing to Unseen Domains: A Survey on Domain Generalization](https://mlsanity.com/paper/2103.03097/) (2021)
- [OGB-LSC: A Large-Scale Challenge for Machine Learning on Graphs](https://mlsanity.com/paper/2103.09430/) (2021)
- [Do Transformers Really Perform Bad for Graph Representation?](https://mlsanity.com/paper/2106.05234/) (2021)
- [Pre-training Molecular Graph Representation with 3D Geometry](https://mlsanity.com/paper/2110.07728/) (2021)
- [Deep Stable Learning for Out-Of-Distribution Generalization](https://mlsanity.com/paper/2104.07876/) (2021)
- [Self-supervised Graph-level Representation Learning with Local and Global Structure](https://mlsanity.com/paper/2106.04113/) (2021)
- [Molecule Attention Transformer](https://mlsanity.com/paper/2002.08264/) (2020)
- [SMILES Transformer: Pre-trained Molecular Fingerprint for Low Data Drug Discovery](https://mlsanity.com/paper/1911.04738/) (2019)
- [Exploiting Neuron and Synapse Filter Dynamics in Spatial Temporal Learning of Deep Spiking Neural Network](https://mlsanity.com/paper/2003.02944/) (2020)

## Built on by (most cited first, 11 of 11)

- [Lo-Hi: Practical ML Drug Discovery Benchmark](https://mlsanity.com/paper/2310.06399/) (2023)
- [mCLM: A Modular Chemical Language Model that Generates Functional and Makeable Molecules](https://mlsanity.com/paper/2505.12565/) (2025)
- [MolClaw: An Autonomous Agent with Hierarchical Skills for Drug Molecule Evaluation, Screening, and Optimization](https://mlsanity.com/paper/2604.21937/) (2026)
- [GeSS: Benchmarking Geometric Deep Learning under Scientific Applications with Distribution Shifts](https://mlsanity.com/paper/2310.08677/) (2023)
- [AMPCliff: quantitative definition and benchmarking of activity cliffs in antimicrobial peptides](https://mlsanity.com/paper/2404.09738/) (2024)
- [SynthFormer: Equivariant Pharmacophore-based Generation of Synthesizable Molecules for Ligand-Based Drug Design](https://mlsanity.com/paper/2410.02718/) (2024)
- [Contextual Molecule Representation Learning from Chemical Reaction Knowledge](https://mlsanity.com/paper/2402.13779/) (2024)
- [Conditional Chemical Language Models are Versatile Tools in Drug Discovery](https://mlsanity.com/paper/2507.10273/) (2025)
- [ATOM: A Pretrained Neural Operator for Multitask Molecular Dynamics](https://mlsanity.com/paper/2510.05482/) (2025)
- [MoleculeCLA: Rethinking Molecular Benchmark via Computational Ligand-Target Binding Analysis](https://mlsanity.com/paper/2406.17797/) (2024)
- [JEDEL: Zero-Shot DNA-Encoded Library Design for Early-Stage Drug Discovery](https://mlsanity.com/paper/2606.23745/) (2026)

## Abstract (arXiv)

Activity cliffs (ACs), which are generally defined as pairs of structurally similar molecules that are active against the same bio-target but significantly different in the binding potency, are of great importance to drug discovery. Up to date, the AC prediction problem, i.e., to predict whether a pair of molecules exhibit the AC relationship, has not yet been fully explored. In this paper, we first introduce ACNet, a large-scale dataset for AC prediction. ACNet curates over 400K Matched Molecular Pairs (MMPs) against 190 targets, including over 20K MMP-cliffs and 380K non-AC MMPs, and provides five subsets for model development and evaluation. Then, we propose a baseline framework to benchmark the predictive performance of molecular representations encoded by deep neural networks for AC prediction, and 16 models are evaluated in experiments. Our experimental results show that deep learning models can achieve good performance when the models are trained on tasks with adequate amount of data, while the imbalanced, low-data and out-of-distribution features of the ACNet dataset still make it challenging for deep neural networks to cope with. In addition, the traditional ECFP method shows a natural advantage on MMP-cliff prediction, and outperforms other deep learning models on most of the data subsets. To the best of our knowledge, our work constructs the first large-scale dataset for AC prediction, which may stimulate the study of AC prediction models and prompt further breakthroughs in AI-aided drug discovery. The codes and dataset can be accessed by https://drugai.github.io/ACNet/.