Table 1 Summary of recent machine learning approaches for odor prediction

Study	Machine learning model	Number of odorants	Feature source	Performance metrics	Reference number
Nozaki, Y. & Nakamoto, T. (2016)	Deep ANN	121	Mass spectral data	R ≈ 0.76	³
Shang, L. et al. (2017)	SVM (Boruta-C), ELM (PCA)	1026	DRAGON Physicochemical Parameters (PCA/Boruta)	SVM Accuracy: 97.08% ELM Accuracy: 97.53 ± 1.35%	⁴
Sharma, A. et al. (2021)	DNN (PPMF), CNN (Xception on 2D images)	5185	PaDel fingerprints; RDKit 2D chemical images	DNN Accuracy: 97.3% CNN Accuracy: 98.3% Combined Precision: 100% (64 smells)	⁵
Saini, K. & Ramanathan, V. (2022)	Daylight-BR	7374	Mordred, Morgan, Daylight fingerprints	micro-F1: 0.3523	⁶
Schicker, D. et al. (2023)	Olfactory Weighted Sum (linear)	64	SMARTS structural patterns	Predicted Accuracy: 0.677 Training Accuracy: 0.905 Random Guessing Performance: 0.214	⁷
Zhang, M. et al. (2024)	Mol-PECO (Deep Learning, Coulomb Matrix/LPE)	8503	Coulomb matrix + LPE encodings	AUROC: 0.813 AUPRC: 0.181	⁸
This study	RF, XGB, LGBM	8681	Morgan fingerprints (ST Model)	XGB-ST AUROC: 0.828 XGB-ST AUPRC: 0.237 LGBM-ST AUROC 0.810 LGBM-ST AUPRC 0.228	Table 2

Overview of major studies applying machine learning (ML) models to olfactory prediction tasks. Shown are the model types, number of odorants used for training, feature sources, and key performance metrics.
ANN artificial neural network, R Pearson correlation coefficient, SVM support vector machine, ELM extreme learning machine, PCA principal component analysis, CNN convolutional neural network, DNN deep neural network, PPMF physiochemical properties and molecular fingerprints, BR binary relevance, SMARTS SMILES arbitrary target specification, LPE learned positional encoding, ST structural (Morgan) fingerprint, AUROC area under the receiver operating characteristic curve, AUPRC area under the precision–recall curve, RF Random Forest, XGB eXtreme Gradient Boosting, LGBM Light Gradient Boosting Machine.

Quick links

Search