DEL-Thyroid: deep ensemble learning framework for detection of thyroid cancer progression through genomic mutation

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Machine Learning in Oncology

Background:

• Genetic mutations are a key factor in cancer development.
• Deep learning models show significant potential for identifying cancer-associated mutations.
• Thyroid cancer is a prevalent malignancy in the USA, necessitating early detection methods.

Purpose of the Study:

• To develop and evaluate an ensemble learning model for early detection of thyroid cancer mutations.
• To leverage deep learning techniques including Long Short-Term Memory (LSTM), Gated Recurrent Units (GRUs), and Bi-directional LSTM (Bi-LSTM).

Main Methods:

• An ensemble deep learning model integrating LSTM, GRUs, and Bi-LSTM was developed.
• The model was trained on a dataset of 633 samples with 969 mutations across 41 genes from asia.ensembl.org and IntOGen.org.
• Feature extraction involved Hahn moments, central moments, raw moments, and matrix-based methods, with evaluation using self-consistency, independent set, and 10-fold cross-validation tests.

Main Results:

• The ensemble learning model achieved 96% accuracy in the independent set test (IST).
• Comprehensive statistical evaluation included metrics like accuracy, recall, sensitivity, specificity, MCC, loss, and F1 Score.

Conclusions:

• The proposed ensemble deep learning model demonstrates high performance in detecting thyroid cancer mutations.
• This approach offers a promising strategy for early identification of genetic alterations linked to thyroid cancer.