MultiCon: A Semi-Supervised Approach for Predicting Drug Function from Chemical Structure Analysis

Abstract

Semi-supervised learning has proved its efficacy in utilizing extensive unlabeled data to alleviate the use of a large amount of supervised data and improve model performance. Despite its tremendous potential, semi-supervised learning has yet to be implemented in the field of drug discovery. Empirical testing of drugs and their classification is costly and time-consuming. In contrast, predicting therapeutic applications of drugs from their structural formulas using semi-supervised learning would reduce costs and time significantly. Herein, we employ a new multicontrastive-based semi-supervised learning algorithmMultiConfor classifying drugs into 12 categories, according to therapeutic applications, on the basis of image analyses of their structural formulas. By rational use of data balancing, online augmentations of the drug image data during training, and the combined use of multicontrastive loss with consistency regularization, MultiCon achieves better class prediction accuracies when compared with the state-of-the-art machine learning methods across a variety of existing semi-supervised learning benchmarks. In particular, it performs exceptionally well with a limited number of labeled examples. For instance, with just 5000 labeled drugs in a PubChem (D3) data set, MultiCon achieved a class prediction accuracy of 97.74%.

Potential diagnosis of COVID-19 from chest X-ray and CT findings using semi-supervised learning

“…For the detection of COVID-19 from the digital chest X-Rays and CT scans, we have used our algorithm, COVIDCon, which is based on our previously developed algorithm [ 49 ]. We have compared the performance of COVIDCon with other state-of-the-art SSL methods including MixMatch [ 10 ], Virtual Adversarial Training (VAT) [ 40 ], Pseudo-labeling [ 32 ], Mean Teacher [ 55 ], model [ 31 ], Interpolation Consistency Training (ICT) [ 60 ] in (Table 2 ).…”

“…Though we applied MultiCon [ 49 ] in the field of drug classification, the use of contrastive learning is still relatively unexplored in the field of medical imaging. We observed that, similar to drug classification, X-ray and CT image classification, especially in the case of COVID-19, is mostly dependent on discerning subtle features.…”

“…Up until now, many SSL models, such as Pseudo-labelling, [ 32 ] VAT, [ 40 ] -Model, [ 31 ] Mean Teacher, [ 55 ] MixMatch, [ 10 ] and FixMatch [ 53 ] have been implemented successfully for analyzing two-dimensional image data. Recently, we also reported a semi-supervised image classification algorithm, MultiCon [ 49 ], which uses multi-contrastive learning alongside consistency regularization to outperform other semi-supervised approaches in image classification.…”

“…Herein, we use the state-of-the-art SSL algorithms and our algorithm, MultiCon [ 49 ], for the classification of grey-scale X-ray and CT scan images as COVID-19, pneumonia, or normal cases from a publicly available COVID-19 radiography dataset [ 47 ] and a COVID-19 CT Scan dataset [ 71 ]. In the present work, we refer MultiCon as COVIDCon for better understanding to the readers.…”

Machine Learning in Chemical Product Engineering: The State of the Art and a Guide for Newcomers

“…The authors in [44] explored semi-supervised variational autoencoders for biomedical relation extraction. Semisupervised learning was also applied in drug discovery for the prediction of drug function from chemical structure analysis [45]. The authors in [46] employed semi-supervised local kernel regression for the soft sensor modeling of the rubber-mixing process.…”