The Construction and Evaluation of a Neural Network-Based Deep Learning Model using Transcriptomic Data to Predict Alzheimer’s Disease-Related Neuropathological Indexes

Session Number

Project ID: CMPS 04

Advisor(s)

Dr. Jubao Duan; NorthShore University HealthSystem

Dr. Siwei Zhang, NorthShore University HealthSystem

Discipline

Computer Science

Start Date

20-4-2022 10:45 AM

End Date

20-4-2022 11:00 AM

Abstract

The elderly population is disproportionately affected by Alzheimer’s disease, which is observed cognitively. Using a set of next-generation RNA sequencing project (ROSMAP) from NorthShore University HealthSystem combined with clinical diagnostic profiles, we constructed a neural network-based deep learning model to predict the occurrence and severity of three geriatric-related pathological features. The final consensus cognitive diagnosis (cogdx) was taken post-mortem. Braak stage is the severity of neuritic tangles (braaksc), and CERAD score measures the number of neuritic plaques (ceradsc). This machine learning module utilized neural networks by using the Python Scikit-Learn package to establish a categorial learning model using the severity evaluation scores as category levels. The model used one-hot encoding at the computational level and two layers of 256 and 64 neurons. Finally, on the evaluation of predication efficiencies, cogdx, which had two categorial levels of 1 and 2, had an accuracy of 99%. The CERAD score also had categorial levels of 1 and 2 and had an accuracy of 96%. Lastly, the Braak set had severity levels from 0 to 3. However, during the subsequent analysis, there was no category 0 presented in the training data, resulting in three categorial levels. The model had an accuracy of 45% .

Share

COinS
 
Apr 20th, 10:45 AM Apr 20th, 11:00 AM

The Construction and Evaluation of a Neural Network-Based Deep Learning Model using Transcriptomic Data to Predict Alzheimer’s Disease-Related Neuropathological Indexes

The elderly population is disproportionately affected by Alzheimer’s disease, which is observed cognitively. Using a set of next-generation RNA sequencing project (ROSMAP) from NorthShore University HealthSystem combined with clinical diagnostic profiles, we constructed a neural network-based deep learning model to predict the occurrence and severity of three geriatric-related pathological features. The final consensus cognitive diagnosis (cogdx) was taken post-mortem. Braak stage is the severity of neuritic tangles (braaksc), and CERAD score measures the number of neuritic plaques (ceradsc). This machine learning module utilized neural networks by using the Python Scikit-Learn package to establish a categorial learning model using the severity evaluation scores as category levels. The model used one-hot encoding at the computational level and two layers of 256 and 64 neurons. Finally, on the evaluation of predication efficiencies, cogdx, which had two categorial levels of 1 and 2, had an accuracy of 99%. The CERAD score also had categorial levels of 1 and 2 and had an accuracy of 96%. Lastly, the Braak set had severity levels from 0 to 3. However, during the subsequent analysis, there was no category 0 presented in the training data, resulting in three categorial levels. The model had an accuracy of 45% .