Scaffold Splits Overestimate Virtual Screening Performance
This work highlights a critical flaw in data splitting methods for virtual screening, impacting drug discovery by showing that current benchmarks may not reflect real-world scenarios, and it is incremental in proposing an alternative approach.
The study found that scaffold splits, commonly used to benchmark AI models in virtual screening, overestimate performance because molecules with different scaffolds can still be similar, leading to unrealistic similarities between training and test sets. Using more accurate UMAP clustering splits resulted in much worse model performance across three AI models on 60 datasets, each with 30,000 to 50,000 molecules.
Virtual Screening (VS) of vast compound libraries guided by Artificial Intelligence (AI) models is a highly productive approach to early drug discovery. Data splitting is crucial for better benchmarking of such AI models. Traditional random data splits produce similar molecules between training and test sets, conflicting with the reality of VS libraries which mostly contain structurally distinct compounds. Scaffold split, grouping molecules by shared core structure, is widely considered to reflect this real-world scenario. However, here we show that the scaffold split also overestimates VS performance. The reason is that molecules with different chemical scaffolds are often similar, which hence introduces unrealistically high similarities between training molecules and test molecules following a scaffold split. Our study examined three representative AI models on 60 NCI-60 datasets, each with approximately 30,000 to 50,000 molecules tested on a different cancer cell line. Each dataset was split with three methods: scaffold, Butina clustering and the more accurate Uniform Manifold Approximation and Projection (UMAP) clustering. Regardless of the model, model performance is much worse with UMAP splits from the results of the 2100 models trained and evaluated for each algorithm and split. These robust results demonstrate the need for more realistic data splits to tune, compare, and select models for VS. For the same reason, avoiding the scaffold split is also recommended for other molecular property prediction problems. The code to reproduce these results is available at https://github.com/ScaffoldSplitsOverestimateVS