Classifying Alzheimer's disease and normal subjects using machine learning techniques and genetic-environmental...

Area of Science:

• Computational biology
• Genetics
• Neuroscience

Background:

• Alzheimer's disease (AD) is a complex neurodegenerative disorder influenced by genetic and environmental factors.
• The diagnostic accuracy of artificial neural networks (ANNs) for AD, considering these multifactorial influences, remains underexplored.

Purpose of the Study:

• To evaluate the accuracy of ANNs in identifying Alzheimer's disease (AD) by integrating common genetic and environmental risk factors.
• To compare the performance of ANNs against other machine learning models like Random Forest (RF) and Support Vector Machine (SVM).

Main Methods:

• A cohort of 184 probable AD patients and 3773 healthy elderly individuals (≥65 years) was analyzed.
• Multilayer ANNs were developed using 51 AD-related single nucleotide polymorphisms (SNPs) and 8 environmental factors as input features.
• Model performance was validated using traditional statistical methods and compared with RF and SVM algorithms.

Main Results:

• The ANN model achieved high accuracy (0.98), sensitivity (0.95), and specificity (0.96) in classifying AD.
• Excluding age and genetic data still resulted in strong performance (accuracy: 0.97, sensitivity: 0.94, specificity: 0.96).
• Feature importance analysis highlighted mental evaluation, education years, and specific SNPs (CASS4 rs7274581, PICALM rs3851179, TOMM40 rs2075650) as crucial for AD prediction.

Conclusions:

• ANN models demonstrate significant accuracy, sensitivity, and specificity for Alzheimer's disease classification.
• Specific genetic variations, including CASS4 rs7274581, PICALM rs3851179, and TOMM40 rs2075650, are important predictors of AD.
• The ANN model's performance is comparable to RF and SVM, confirming its utility in managing the complexity of AD.