Graph Neural Networks for Identifying Protein-Reactive Compounds
Bibliographic record
Abstract
The identification of protein-reactive electrophilic compounds is critical to the design of new covalent modifier drugs, screening for toxic compounds, and the exclusion of reactive compounds from high throughput screening. In this work, we employ traditional and graph machine learning (ML) algorithms to classify molecules being reactive towards proteins or nonreactive. For training data, we built a new dataset, ProteinReactiveDB, composed primarily of covalent and noncovalent inhibitors from the DrugBank, BindingDB, and CovalentInDB databases. To assess the transferability of the trained models, we created a custom set of covalent and noncovalent inhibitors, which was constructed from the recent literature. Baseline models were developed using Morgan fingerprints as training inputs, but they performed poorly when applied to compounds outside the training set. We then trained various Graph Neural Networks (GNNs), with the best GNN model achieving an Area Under the Receiver Operator Characteristic (AUROC) curve of 0.80, precision of 0.89, and recall of 0.72. We also explore the interpretability of these GNNs using Gradient Activation Mapping (GradCAM), which shows regions of the molecules GNNs deem most relevant when making a prediction. These maps indicated that our trained models can identify electrophilic functional groups in a molecule and classify molecules as protein-reactive based on their presence. We demonstrate the use of these models by comparing their performance against common chemical filters, identifying covalent modifiers in the ChEMBL database and generating a putative covalent inhibitor based on an established noncovalent inhibitor.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".