Post

Christian Dallago
@machine.learning.bio

Five years ago, we released FLIP. The core question was: can ML models for protein fitness prediction generalize in the ways that actually matter for protein engineering, i.e. low data, extrapolation to more mutations, out-of-distribution sequences?

5 reposts4 likes

Jeremy Nelson
@jeremynelson.bsky.social

FLIP was groundbreaking work. The extrapolation challenge in protein engineering mirrors what we see in production ML systems—models that score well on benchmarks but fail on edge cases. Have you explored whether ensembling with uncertainty quantification helps with those low-data regime prediction…

Christian Dallago
@machine.learning.bio

FLIP spawned fast development of several different benchmarking efforts across protein design, engineering, and variant effect assesment. The answer in 2021 was: sometimes, but simpler models hold up surprisingly well.

Christian Dallago
@machine.learning.bio

We were interested in how things had changed 5 years after our first release. So, we built FLIP2 on select datasets from great labs across the world, many of which have gracefully agreed to make their data freely available.

Christian Dallago
@machine.learning.bio

FLIP2, adds seven new sequence-fitness landscapes - industrial enzymes, nucleases, rhodopsins, protein-protein interactions - and 16 splits that test the generalization axes protein engineers really hit: more mutations, new positions, higher fitness, different wild-types.

Christian Dallago
@machine.learning.bio

The answer in 2026 is largely the same. Simple ridge regression on one-hot sequences, optionally supplemented with zero-shot pLM likelihoods, often matches or outperforms fine-tuned protein language models.