Guide your favorite protein sequence generative model

Generative machine learning models have begun to transform protein engineering, yet no principled framework for conditioning on auxiliary information in a plug-and-play manner exists; one may want to iteratively incorporate experimental feedback, or make use of an existing classifier -- such as for predicting enzyme commission number -- in order to guide the sampling of the generative model to generate sequences with desired properties. Herein, we present ProteinGuide, a rigorous and general framework to achieve just that: through unifying a broad class of protein generative models that includes masked language, (order-agnostic) autoregressive, diffusion and flow-matching models, we provide an approach to statistically condition pre-trained protein generative models. We demonstrate applicability of our approach by guiding each of two commonly used protein generative models, ProteinMPNN and ESM3, to generate amino acid and structure token sequences conditioned on several user-specified properties, namely, enhanced stability and CATH-labeled fold generation.

ColorSwap: A Color and Word Order Dataset for Multimodal Evaluation

This paper introduces the ColorSwap dataset, designed to assess and improve the proficiency of multimodal models in matching objects with their colors. The dataset is comprised of 2,000 unique image-caption pairs, grouped into 1,000 examples. Each example includes a caption-image pair, along with a ``color-swapped'' pair. We follow the Winoground schema: the two captions in an example have the same words, but the color words have been rearranged to modify different objects. The dataset was created through a novel blend of automated caption and image generation with humans in the loop. We evaluate image-text matching (ITM) and visual language models (VLMs) and find that even the latest ones are still not robust at this task. GPT-4V and LLaVA score 72% and 42% on our main VLM metric, although they may improve with more advanced prompting techniques. On the main ITM metric, contrastive models such as CLIP and SigLIP perform close to chance (at 12% and 30%, respectively), although the non-contrastive BLIP ITM model is stronger (87%). We also find that finetuning on fewer than 2,000 examples yields significant performance gains on this out-of-distribution word-order understanding task. The dataset is here: https://github.com/Top34051/colorswap.