Machine learning random forest for predicting oncosomatic variant NGS analysis.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Next-generation sequencing (NGS) generates vast amounts of data for cancer diagnosis and treatment.
• Analyzing NGS variants is time-consuming and prone to errors, necessitating advanced analytical tools.
• Artificial intelligence (AI) offers potential solutions for processing complex genomic data.

Purpose of the Study:

• To investigate machine learning (ML) algorithms for classifying NGS variants.
• To develop and implement an ML tool for routine use in cancer diagnosis.
• To improve the accuracy and efficiency of variant analysis in clinical settings.

Main Methods:

• Compared various ML algorithms, including random forest and neural networks, using k-fold cross-validation.
• Trained ML models on a local database of 102,011 variants with 7 parameters.
• Validated models using 30% of the data, focusing on accuracy and error rates.

Main Results:

• Random forest (RF) achieved an error rate of 0.22% with an AUC of 0.99.
• Neural networks demonstrated 98% accuracy and an ROC-AUC of 0.99.
• Routine implementation of the RF model reduced the error rate to below 0.5%.

Conclusions:

• Random forest is a highly accurate and efficient ML model for classifying NGS variants in cancer.
• AI, particularly RF, can be routinely implemented to assist biologists in NGS data interpretation.
• ML models show promise for enhancing the accuracy of cancer diagnostics and potentially predicting complex variants.