Integration of Random Forest Classifiers and Deep Convolutional Neural Networks for Classification and Biomolecular Modeling of Cancer Driver Mutations

Abstract

Development of machine learning solutions for prediction of functional and clinical significance of cancer driver genes and mutations are paramount in modern biomedical research and have gained a significant momentum in a recent decade. In this work, we integrate different machine learning approaches, including tree based methods, random forest and gradient boosted tree (GBT) classifiers along with deep convolutional neural networks (CNN) for prediction of cancer driver mutations in the genomic datasets. The feasibility of CNN in using raw nucleotide sequences for classification of cancer driver mutations was initially explored by employing label encoding, one hot encoding, and embedding to preprocess the DNA information. These classifiers were benchmarked against their tree-based alternatives in order to evaluate the performance on a relative scale. We then integrated DNA-based scores generated by CNN with various categories of conservational, evolutionary and functional features into a generalized random forest classifier. The results of this study have demonstrated that CNN can learn high level features from genomic information that are complementary to the ensemble-based predictors often employed for classification of cancer mutations. By combining deep learning-generated score with only two main ensemble-based functional features, we can achieve a superior performance of various machine learning classifiers. Our findings have also suggested that synergy of nucleotide-based deep learning scores and integrated metrics derived from protein sequence conservation scores can allow for robust classification of cancer driver mutations with a limited number of highly informative features. Machine learning predictions are leveraged in molecular simulations, protein stability, and network-based analysis of cancer mutations in the protein kinase genes to obtain insights about molecular signatures of driver mutations and enhance the interpretability of cancer-specific classification models.

Predicting long-term multicategory cause of death in patients with prostate cancer: random forest versus multinomial model

“…Despite the slightly better accuracy linked to data with one-hot encoding than standard encoding, we found no statistical differences between the two methods. This finding is inconsistent with previous reports,[28, 29] and needs further validation. We also noticed that the minimal depths of trees in our best-fit RF models were usually 1 to 3.…”

“…[28] This approach was also used in machine learning models of cancer driver genes. [29] For one-hot encoding, all multicategory variables (i.e. discrete variables with more than two catrgories) were transformed into a new set of binary variables.…”

Multiscale modeling: foundations, historical milestones, current status, and future prospects

“…This success notwithstanding, the design of new MSM algorithms in coupling different scales, data utilization, and their implementation on HPC is becoming increasingly cumbersome in the face of heterogeneous data availability and rapidly evolving HPC architectures and platforms. On the other hand, while purely datadriven models of molecular and cellular systems spawned by the techniques of data science [132][133][134], and in particular, ML methods including deep learning methods [135][136][137], are easy to train and implement, the underlying model manifests as a black-box. This general approach taken by the ML community is well suited for classification, learning, and regression problems, but suffers from limitations in interpretability and explainability, especially when mechanism-based understanding is a primary goal.…”

DeepSnap-Deep Learning Approach Predicts Progesterone Receptor Antagonist Activity With High Performance

“…In addition, the prediction results of the current model are often difficult to develop the drug discovery for clinical trials (Gayvert et al, 2016;Neves et al, 2018;Vamathevan et al, 2019). A deep learning (DL) approach with convolutional neural networks (CNNs), Rectified Linear Unit (ReLU), and max pooling is a promising, powerful tool for the classification modeling (Date and Kikuchi, 2018;Öztürk et al, 2018;Wang et al, 2018;Agajanian et al, 2019;Idakwo et al, 2019;Jo et al, 2019), where factors affecting its prediction performance include sufficient size, suitable representation, and accurate labeling of supervised input datasets (Bello et al, 2019;Chauhan et al, 2019;Liu P. et al, 2019). To resolve these issues, the DL-based QSAR modeling approach using molecular images produced by 3D chemical structure as input data was previously developed and referred to as the DeepSnap-DL approach (Uesawa, 2018).…”