Lemna Bio (lemna.bio), an #EPFL spin-off building structural AI models that reason about how molecules interact, directly from 3D structures, is looking for a (paid!) intern to benchmark their newly developed models.
-
Lemna Bio (lemna.bio), an #EPFL spin-off building structural AI models that reason about how molecules interact, directly from 3D structures, is looking for a (paid!) intern to benchmark their newly developed models. Details in the reply to this toot; DM me if interested, feel free to pass it along your network if you know someone who might be interested.
Please only candidates already based in Switzerland.
#getfedihired #switzerland #bioinformatics #StructuralBiology
-
Lemna Bio (lemna.bio), an #EPFL spin-off building structural AI models that reason about how molecules interact, directly from 3D structures, is looking for a (paid!) intern to benchmark their newly developed models. Details in the reply to this toot; DM me if interested, feel free to pass it along your network if you know someone who might be interested.
Please only candidates already based in Switzerland.
#getfedihired #switzerland #bioinformatics #StructuralBiology
Internship project: Benchmarking PeSTo Screen against ATOMICA for molecular interaction networks
About us
Lemna Bio is an EPFL spin-off building structural AI models that reason about how molecules interact, directly from 3D structure. Our published model PeSTo predicts interaction interfaces on proteins. We are now extending it into PeSTo Screen, a model that performs all-against-all interaction screening: given large sets of molecules, it scores how likely each pair is to interact, in a single fast pass rather than by brute-force docking.The project
A recent model from the Zitnik lab at Harvard, ATOMICA, learns representations of molecular interfaces and uses them to build interaction networks and recover disease-associated pathways. ATOMICA connects proteins by how similar their interfaces look. PeSTo Screen takes a different route: it predicts whether two proteins actually interact. This project asks a clear question: does predicting interactions directly recover biology better than comparing interface similarity?
You will benchmark PeSTo Screen head to head against ATOMICA, using ATOMICA’s own published analyses as the yardstick.#gefedihired #switzerland #epfl #bioinformatics #StructuralBiology #StructuralBiology
-
Internship project: Benchmarking PeSTo Screen against ATOMICA for molecular interaction networks
About us
Lemna Bio is an EPFL spin-off building structural AI models that reason about how molecules interact, directly from 3D structure. Our published model PeSTo predicts interaction interfaces on proteins. We are now extending it into PeSTo Screen, a model that performs all-against-all interaction screening: given large sets of molecules, it scores how likely each pair is to interact, in a single fast pass rather than by brute-force docking.The project
A recent model from the Zitnik lab at Harvard, ATOMICA, learns representations of molecular interfaces and uses them to build interaction networks and recover disease-associated pathways. ATOMICA connects proteins by how similar their interfaces look. PeSTo Screen takes a different route: it predicts whether two proteins actually interact. This project asks a clear question: does predicting interactions directly recover biology better than comparing interface similarity?
You will benchmark PeSTo Screen head to head against ATOMICA, using ATOMICA’s own published analyses as the yardstick.#gefedihired #switzerland #epfl #bioinformatics #StructuralBiology #StructuralBiology
What you will do
1. Run PeSTo Screen all-against-all over a large set of human proteins to generate a predicted interaction network.
2. Deploy and characterize the results against reference interaction databases such as STRING, to see how well the predictions line up with known biology.
3. Recreate ATOMICA’s disease-network analysis on your PeSTo Screen network, overlay disease-associated genes, and test which approach recovers disease pathways more cleanly. Comparison to ATOMICA is the core deliverable.Optional extension
If time allows and depending on model scope, benchmark PeSTo Screen on standardized external tasks such as the RNAGlib RNA structure-function benchmark, and other.#gefedihired #switzerland #epfl #bioinformatics #StructuralBiology #StructuralBiology
-
What you will do
1. Run PeSTo Screen all-against-all over a large set of human proteins to generate a predicted interaction network.
2. Deploy and characterize the results against reference interaction databases such as STRING, to see how well the predictions line up with known biology.
3. Recreate ATOMICA’s disease-network analysis on your PeSTo Screen network, overlay disease-associated genes, and test which approach recovers disease pathways more cleanly. Comparison to ATOMICA is the core deliverable.Optional extension
If time allows and depending on model scope, benchmark PeSTo Screen on standardized external tasks such as the RNAGlib RNA structure-function benchmark, and other.#gefedihired #switzerland #epfl #bioinformatics #StructuralBiology #StructuralBiology
Who we are looking for
You are finishing or have recently finished an MSc in computational biology, bioinformatics, or a related field. What matters most is that you have a good understanding of biological networks, and good computational skills.
Required
• Strong Python and the scientific stack (NumPy, pandas)
• Experience handling protein structure data (PDB/mmCIF, AlphaFold DB) and building reproducible data pipelines over thousands of structures
• Working knowledge of network and graph analysis (NetworkX, igraph, clustering and community detection)
• Familiarity with biological interaction and annotation resources (STRING, BioGRID, UniProt, disease gene sets such as OpenTargets or DisGeNET)
• Benchmarking rigor: you understand precision/recall trade-offs, how to build fair negative sets, and how to spot circularity
• Ability to read a published method and reproduce its analysis from paper plus code
• Good scientific writing and communication in EnglishDuration: 6 months
Paid: CHF 2,200 gross monthly#gefedihired #switzerland #epfl #bioinformatics #StructuralBiology #StructuralBiology
-
F folfdk@helvede.net shared this topic
