Introduction - Willing To Contribute (GSOC'26)

[vinitjain2005]

Hi @jonbrenas @tristanpwdennis,

My name is Vinit Jain, and I’m a 3rd year student currently focusing on Python and machine learning, with a growing interest in bioinformatics applications.

I’ve been exploring the Building a machine-learning taxon classifier for genomic classification in malaria mosquitoes project and found the problem of FASTQ-based classification particularly interesting—especially the idea of avoiding full variant calling pipelines and working directly with raw sequencing reads.

I’ve started going through the malariagen-data-python repository to better understand the existing API and data workflows, and I’m currently identifying areas where I can make a meaningful first contribution. I’m also beginning to explore approaches such as k-mer based representations and lightweight classification models as a potential direction for the project.

I plan to submit a PR soon and would really appreciate any guidance on areas where contributions would be most valuable for new contributors.

Looking forward to learning and contributing!

Thanks,
Vinit Jain

Github: @vinitjain2005

[jonbrenas]

Hi Vinit,

Thank you for your help. I think you have found ways to contribute without my help.

[sneha11naidu]

Hi everyone,

I’m a Master’s student in Data Science at City, University of London, and I’m planning to apply to GSoC 2026 for the project “Building a machine-learning taxon classifier for genomic classification in malaria mosquitoes.”

From the project description, I understand that the goal is to classify mosquito samples directly from raw FASTQ sequencing data without relying on full variant-calling pipelines.

I’m currently trying to understand the best way to represent raw DNA sequences as input features for machine learning models.

I wanted to ask:

• Are there any preferred approaches within MalariaGEN for converting sequencing reads into features for classification?
• Should the focus be on simpler representations (e.g. frequency-based features) to keep the model lightweight?

I’d really appreciate any guidance on aligning the approach with existing workflows.

Thank you!

[jonbrenas]

Hello @sneha11naidu,

Thank you for your interest. There are no preferred approaches to converting sequencing reads for classification, we have not done it in the past. Ideally, the model would be as lightweight as possible but there is certainly a right balance between memory or computation needs and accuracy.

Best,
Jon

[sneha11naidu]

Hi Jon,

Thank you for your response, that’s very helpful.

I’ll focus on designing a lightweight approach while balancing accuracy, likely starting with simple feature representations and evaluating their effectiveness before considering more complex methods.

I appreciate the guidance — I’ll incorporate this into my proposal.

Best,
Sneha