Genetic Variant Classifier

Quick video link: https://app.screencastify.com/v2/watch/je3gP3mNTPuhsQHmWi4s

Inspiration

Andrea works on rare disease and genetics initiatives in her free time, and being around patients and scientists in this space made her know that this was an important problem to solve with AI. Many genetic variants are still unknown today, and we hope this app will help researchers narrow down the next variants to research.

What it does

Users can input variant data in CSV format, and our app will return if the variants are pathogenic or not.

How we built it

We trained a model on genetic data for CFTR gene using a dataset (Clinvar) containing labelled data with pathogenic or not, and merged it with aggregate data from Gnomad to get extra information about the variants. All this was done with back and forth with ChatGPT, modifying the canvas. Also used this process to get unclassified data to demo with and test.

The UI was written by lovable.

Challenges we ran into

We had trouble merging the datasets because the format of data in one column was different. We had to convert them to compatible formatting using normalization of one of the datasets.

Accomplishments that we're proud of

Getting the model to accurately classify data and showing that it worked on unclassified data.

What we learned

So much :D this was Andrea's first time training a model and both of our first time doing work with genetic data.

What's next for Genetic Variant Classifier

Seeing if we can improve the accuracy even more, trying different models that are more robust (we are using a very simple classifier). Expanding to different genes such as those associated with cancer.