Felipe Colombelli

Hey there, welcome to my—hopefully lighthearted—profile! Drop the formalities and let's talk science.

I'm a PhD student at KTH in the EECS school, based at SciLifeLab. At the moment, I’m particularly excited about multimodal deep learning for spatial transcriptomics—building models that integrate histology, spatial gene expression, and related omics to improve biological insight and prediction.

I've been sciencing around for some time, both in academia in Brazil—at the Bioinformatics Core of HCPA—and with industry partners—at Dell Technologies and LanguageTool.

When it comes to bioinfo, I've worked with (ensemble) feature selection for biomarker discovery in high-dimensional transcriptomics data—check out my paper published in Knowledge-Based Systems during my undergraduate studies—GNNs for cancer staging, gene co-expression network analysis, and a little bit of genetic variant analysis, as part of a very cool project with a former collaborator interested in CRISPR-based therapies.

As per my industry adventures, I had the pleasure of exploring a range of topics with different perspectives. These include multiobjective optimization for databases security, chatbots development, and many NLP tasks such as style-transfer rewriting, grammatical error correction, and writing scorers. In the NLP landscape, I mostly worked with LLMs and related techniques such as LoRA fine-tuning, model quantization, knowledge distillation and bayesian neural nets for uncertainty quantification.

That said, I really thrive when I'm able to apply all the AI jazz I've been learning during these past 5+ years to problems in the life sciences.

You're very welcome to write me if you'd like to discuss about a cool idea, propose a collab, ask about my previous work, or anything related to academia. And if you do, please send your message to my SciLifeLab @—which I treat with a bit more love: felco@scilifelab.se