We've been working on ways to predict protein-protein interactions when you don't have massive datasets. Turns out, transfer learning from general molecular representations helps more than expected. Here's what worked and what didn't.
Read Full Article
Real-world sequencing data is messy. We've been testing different normalization approaches and found some surprising results about batch effects.
Continue reading
Metabolic networks are naturally graph-structured. We applied GNNs to pathway analysis and got some interesting insights about substrate specificity.
Continue readingClinical variant interpretation needs both computational power and domain expertise. Our collaboration with medical geneticists taught us a lot about bridging that gap.
Continue readingThoughts from our team on where bioinformatics and AI are heading, and the challenges we're actually facing in practice.
Everyone talks about AI revolutionizing drug discovery. And yes, there's real progress. But what people don't discuss enough is the validation challenge. You can generate thousands of promising candidates computationally, but then you hit the bottleneck of experimental validation. The models are getting better at predicting binding affinity, but biological activity in living systems? That's still incredibly complex. We're making headway, but it's careful, methodical work — not the overnight transformation that headlines suggest.
Open research benefits everyone. When we publish methods, we include our full preprocessing pipeline. Reproducibility matters more than competitive advantage.
Fast models are great for screening, but when it comes to clinical decisions, we often choose slower, more interpretable approaches. Context matters.
We've seen better results with 5,000 carefully curated samples than 50,000 noisy ones. Spending time on data curation pays off.
Hands-on session covering CNNs for sequence analysis, attention mechanisms for variant calling, and transfer learning strategies. Bring your own data if you want specific feedback.
Dr. Lindström presents recent work on applying graph neural networks to metabolic pathway analysis. Discussion and Q&A session follows.
Undergraduate and graduate students interested in computational biology are welcome to visit, meet the team, and see our workflows. Informal setting, lots of questions encouraged.