Apply for a GPU community grant: Academic project
#1
by Geonomic - opened
Project Title: The Genomic Oracle - Cascading AI Pipeline for Genomic Sequence Classification
GPU Justification: I am a bioinformatics graduate student at the University of Maryland Global Campus (UMGC), my team is developing the "Genomic Oracle" for our final capstone project under the sponsorship of the UMGC Bioinformatics Department and Dr. Wolfgang Rumpf.
Our application is a complex, automated machine-learning pipeline that routes raw NGS DNA sequences through a branching decision tree of four integrated models:
- Scikit-Learn Logistic Regression model (Coding vs. Non-Coding).
- fine-tuned DNABERT-2 deep neural network (Promoter vs. Non-Promoter).
- multi-class sequence classifier for functional annotation (e.g., Exons, RNA transcripts).
- custom fine-tuned BERT model utilizing ALiBi positional embeddings to predict phenotypic associations (Lean vs. Obese genes).
There is also an integrated Biopython script that queries the NCBI API to fetch real-world spatial mapping coordinates.
We are currently hitting severe Out of Memory (OOM) crashes on the default 16GB CPU Basic tier. The overlapping k-mer tokenization strategies and the active memory required to hold multiple transformer-based architectures simultaneously exceed standard RAM capabilities.
A Community GPU grant (such as a T4 Small) would provide the necessary VRAM and hardware acceleration to stabilize the cascading handoffs between our models, ensuring fast, reliable inference for our final capstone presentation and public deployment.
Thank you for your consideration.