Leveraging 13C NMR spectroscopic data derived from SMILES to predict the functionality of small biomolecules by machine learning: a case study on human Dopamine D1 receptor antagonists

This study contributes to ongoing research which aims to predict small biomolecule functionality using Carbon-13 Nuclear Magnetic Resonance ($^{13}$C NMR) spectrum data and machine learning (ML). The approach was demonstrated using a bioassay on human dopamine D1 receptor antagonists. The Simplified Molecular Input Line Entry System (SMILES) notations of compounds in this bioassay were extracted and converted into spectroscopic data by software designed for this purpose. The resulting data was then used for ML with scikit-learn algorithms. The ML models were trained by 27,756 samples and tested by 5,466. From the estimators K-Nearest neighbor, Decision Tree Classifier, Random Forest Classifier, Gradient Boosting Classifier, XGBoost Classifier, and Support Vector Classifier, the last performed the best, achieving 71.5 % accuracy, 77.4 % precision, 60.6% recall, 68 % F1, 71.5 % ROC, and 0.749 cross-validation score with 0.005 standard deviation. The methodology can be applied to predict any functionality of any compound when relevant data are available. It was hypothesized also that increasing the number of samples would increase accuracy. In addition to the SMILES $^{13}$C NMR spectrum ML model, the time- , and cost-efficient CID_SID ML model was developed. This model allows researchers who have developed a compound and obtained its PubChem CID and SID to check whether their compound is also a human dopamine D1 receptor antagonist based solely on the PubChem identifiers. The metrics of the CID_SID ML model were 80.2% accuracy, 86.3% precision, 70.4% recall, 77.6% F1, 79.9% ROC, five-fold cross-validation score of 0.8071 with 0.0047 Standard deviation.