Floris van der Flier @florisvdf.bsky.social · 29/06/2026In summary - Uncertainty-aware acquisition can yield better identification of hit variants - Favoring increased model uncertainty is preferred for single mutants - Matching objectives to downstream use cases is worth exploring, even when traditional approaches dominate on average 000
Floris van der Flier @florisvdf.bsky.social · 29/06/2026This matters, because practitioners don't need general-purpose models, they need models that excel on their dataset 100
Floris van der Flier @florisvdf.bsky.social · 29/06/2026Key findings: Traditional regression models still outperform preferential models on most datasets But: Uncertainty-aware acquisition functions consistently win, especially when acquiring in high-uncertainty regions Preferential models show clear advantages on specific datasets 100
Floris van der Flier @florisvdf.bsky.social · 29/06/2026We built a retrospective evaluation protocol using quantile cross-validation (holding out high-value variants) and a custom recovery metric (measuring how well models + acquisition functions prioritize those held-out variants). 100
Floris van der Flier @florisvdf.bsky.social · 29/06/2026We tailored model development and evaluation to match the real objective of acquisition. We tested two techniques: Preferential learning → mirror the goal of variant selection Uncertainty quantification → provide information about model confidence to guide acquisition 100
Floris van der Flier @florisvdf.bsky.social · 29/06/2026On top of that, models are often inaccurate on new data, yet our acquisition strategies ignore where models fail. Model uncertainty can reveal these failure modes and guide smarter variant selection. 100
Floris van der Flier @florisvdf.bsky.social · 29/06/2026Protein engineering models are built to predict exact properties, but when we use them, we only care if a variant is better than what we have. We evaluate them on ranking every variant in a dataset, but we only need them to find the rare, high-performing ones. 100
Floris van der Flier @florisvdf.bsky.social · 29/06/2026We just put out a new preprint! Preferential learning + uncertainty quantification for better protein engineering, check it out here: doi.org/10.64898/202...doi.orgAcquiring Improved Protein Variants With Probabilistic Preferential LearningVariant effect prediction (VEP) models can be used to select promising novel enzymes from a pool of candidates. Most supervised VEP models are framed as regression tasks, placing more emphasis on gett... 150
Floris van der Flier @florisvdf.bsky.social · 13/03/2026We saw similar effects in our study, even for a combinatorial dataset we specifically designed to relate structural characteristics of mutations to their predictability. An augmented Potts model and a PLM based model achieved only marginal gains: doi.org/10.1016/j.cs...doi.orgRedirecting 021
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(8/8) Code is available at github.com/florisvdf/ch....github.comGitHub - florisvdf/chargenet: Predicting protein variant properties using electrostatic representationsPredicting protein variant properties using electrostatic representations - florisvdf/chargenet 000
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(7/8) Despite the outcome, this was still a very fun project to work on. Many thanks my supervisors, Henning Redestig and Dick de Ridder, for their excellent guidance, and Luis-Cascao Pereira and David Estell for inspiring us to explore electrostatic quantities of proteins. 100
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(6/8) We suspect the limitations arise from imperfect computational tools, missing biological context (e.g., post-translational modifications, molecular crowding), and treating proteins as static objects. Still, we believe these results could help narrow the search space for future VEP strategies. 100
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(5/8) We trained and evaluated ChargeNet on datasets from ProteinGym, comparing it to evolutionary models both standalone and in ensemble. Across all tests, ChargeNet offered no advantage—suggesting its physical representations don’t add information beyond what evolutionary models already encode. 100
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(4/8) We tried to design a representation that is evolution agnostic and instead relies on the physical properties. This led to ChargeNet—a pipeline that predicts a variant’s structure with FoldX, computes its 3D electrostatic profile with APBS, and feeds that into a 3D CNN for property prediction. 100
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(3/8) SotA VEP models are evolutionary and excel at spotting “unnatural” mutations, which often correlate with loss of function. However, by pretraining only on natural sequences, such models may miss out on mutations that wouldn’t occur in nature, but still lead to interesting changes in function. 100
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(2/8) Unfortunately, our strategy did not yield the results we hoped for; our model is outperformed by existing models in every scenario we tested. Nevertheless, we find it important to share our findings so that others can build on them, refine their approach, or take more promising directions. 120
Floris van der Flier @florisvdf.bsky.social · 14/08/2025(1/8) Happy to share that our preprint titled “Predicting protein variant properties using electrostatics” is now online! We tested a physics-based approach to variant effect prediction (VEP). www.biorxiv.org/content/10.1...biorxiv.orgPredicting protein variant properties with electrostatic representationsDoes evolution capture the full functional potential of proteins, or is this potential restricted by selective pressures? If the former is true, providing variant effect prediction (VEP) models with e... 130
Floris van der Flier @florisvdf.bsky.social · 03/02/2025Question: Could the experimental uncertainty arise from the dynamics of the protein in solution? And if so, are there examples where this uncertainty strongly correlates with dynamics? 110
Reposted by Floris van der FlierDaniel Probst @skepteis.bsky.social · 16/12/2024If you're looking for a (fully funded) PhD position in #ml / #ai methodology (with plenty of opportunities to apply it) at the intersection of #chemistry and #biology, don't forget to apply before the end of the year. www.wur.nl/en/vacancy/p...wur.nlPhD student - Machine learning in bio- and cheminformatics 12517
Floris van der Flier @florisvdf.bsky.social · 06/12/2024Once again looking amazing. Are you considering at some point curating a gallery to showcase what's possible with molecular nodes? Or does this already exist and did I miss it? 🫣 110
Reposted by Floris van der FlierDaniel Probst @skepteis.bsky.social · 02/12/2024Here's a first: I'm hiring a PhD student. If you have (or know someone who has) a strong background and interest in computational (bio)chemistry or computer science (especially 3D computer vision), let's work together 😊 www.wur.nl/nl/vacature/...wur.nlPhD student - Machine learning in bio- and cheminformatics 04322
