Early Risk Prediction of Chronic Myeloid Leukemia with Protein Sequences using Machine Learning-based Meta-Ensemble

TL;DR

This study presents a machine learning meta-ensemble approach using protein sequences to predict early risk of Chronic Myeloid Leukemia, aiming to improve early diagnosis and treatment outcomes.

Contribution

It introduces a novel multi-layer perceptron-based meta-ensemble system leveraging protein sequence features for early CML risk prediction.

Findings

Achieved improved classification accuracy over existing methods

Utilized amino acid property-based features for effective prediction

Demonstrated potential as a biomarker for early CML diagnosis

Abstract

Leukemia, the cancer of blood cells, originates in the blood-forming cells of the bone marrow. In Chronic Myeloid Leukemia (CML) conditions, the cells partially become mature that look like normal white blood cells but do not resist infection effectively. Early detection of CML is important for effective treatment, but there is a lack of routine screening tests. Regular check-ups and monitoring of symptoms are the best way to detect CML in the early stages. In the study, we developed a multi-layer-perception-based meta-ensemble system using protein amino acid sequences for early risk prediction of CML. The deleterious mutation analysis of protein sequences provides 7discriminant information in amino acid sequences causing CML. The protein sequences are expressed into molecular descriptors using the values of hydrophobicity and hydrophilicity of the amino acids. 9 These descriptors are transformed in various statistical and correlation-based feature spaces. These 10 features information is given to several diverse types of base learners. The preliminary predictions of 11 base-learners are employed to develop Multi-Layered Perceptron (MLP) based meta-ensemble. The 12 proposed learning approach effectively utilizes the discriminant information to classify CML/non- 13 CML protein sequences. The proposed prediction system has given improved results and it can be 14 employed as a potential biomarker for early diagnosis of CML.