1 of 11

Introduction to WeeklyPulse: HealthPredict Epigenetics Prediction Application

Model training and data entry demonstration

Presenters: Jordan Bahnub, Runhao Wang

2 of 11

Current Treatments and Technology

Reactive, focus on treatment after disease reaches symptom stage
Treatment incurs high cost and low survival
DNA sequence only predicts risk, not current disease
Traditional biomarkers are complicated and expensive

But what if all these issues could be fixed? What if the data from a single drop of blood could be enough to get a comprehensive view of the patient?

We don’t have to wonder ‘what if’ anymore.

3 of 11

HealthPredict: A method for diagnosis, risk, and prevention--all in one tool

Data-driven prediction

Using a large number of medical data, through machine learning model analysis, key features are extracted to achieve early prediction of cancer risk and provide more accurate prevention strategies for patients.

The Application of Deep Learning in Disease Screening

Deep learning technology, with elastic convolutional neural networks (CNN), was used to analyze RNA-seq expression and methylation data to, improve the accuracy and efficiency of diagnostic screening to assay cancer risk and biological health.

Early warning system

Data from epigenetic blood tests integrated to both further validate and train the AI while providing patient diagnostics, detecting disease long before it occurs.

4 of 11

Ideas and methods

Data-driven prediction model that learns and grows

A large number of historical medical data are used to train the model, and the past health records and living habits of patients are analyzed to provide a scientific basis for disease prediction and risk assessment.

Health-monitoring over time

Blood samples will be assayed for epiegetic features like methylation, histone acetylation, and chromatin structure. Results will be integrated into a website to track data over time.

Individualized treatment plan design

According to the patient's epigenetic information, disease type and development stage, the personalized treatment plan is customized to improve the treatment effect and reduce side effects.

5 of 11

Modeling demo

6 of 11

Train and export CNN in minutes with settings of your choice.

7 of 11

Cancer data from TCGA or personal datasets can be analyzed for genetic heatmapping.

8 of 11

The training model features…

Continuous optimization and learning

With the accumulation of more data, the model will continue to self-optimize and learn to improve the prediction accuracy and adapt to changing medical needs.

User data training

Users can directly upload large batches of .csv file data to train the website internal model, then export the model in the file format of their choice.

Hundreds of genes and methylation sites with conversions to common names

9 of 11

Next Steps…

This is only the beginning

If this could all be integrated together within 24 hours, imagine what we could do with months of thorough refinement.

We plan to make more expression heatmaps for common diseases and conditions, and integrate diagnostics together into one upload tab that gives comprehensive analysis of multiple different conditions at once.

We will continue to add data and create better standardization procedures, to make it simple, easy, and efficient for users to output results.