Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Steering Protein Family Design through Profile Bayesian Flow
Authors: Jingjing Gong, Yu Pei, Siyu Long, Yuxuan Song, Zhe Zhang, Wenhao Huang, Ziyao Cao, Shuyi Zhang, Hao Zhou, Wei-Ying Ma
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results show that Profile BFN has a profound understanding of proteins. When generating diverse and novel family proteins, it can accurately capture the structural characteristics of the family. |
| Researcher Affiliation | Academia | Jingjing Gong1 Yu Pei1 Siyu Long1 Yuxuan Song1 Zhe Zhang1 Wenhao Huang1 Ziyao Cao1 Shuyi Zhang2 Hao Zhou1 Wei-Ying Ma1 1 Institute of AI Industry Research (AIR), Tsinghua University 2 School of Pharmaceutical Sciences, Tsinghua University EMAIL EMAIL |
| Pseudocode | Yes | B ALGORITHMS Algorithm 1 Training Loss Procedure Algorithm 2 Family Protein Generation Procedure |
| Open Source Code | No | The paper does not provide an explicit statement about releasing the source code for the methodology described, nor does it include a link to a code repository. |
| Open Datasets | Yes | Training Dataset In line with ESM-2, we use protein sequence data from the Uni Ref database (Suzek et al., 2007) (as of March 2024) to train Profile BFN. |
| Dataset Splits | No | The paper uses datasets such as UniRef, CAMEO, and enzyme families but does not explicitly provide specific training, validation, or test split percentages or sample counts for these datasets in the main text or appendices. |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware (e.g., GPU models, CPU types) used for running the experiments. |
| Software Dependencies | No | The paper mentions using tools and libraries such as AdamW, HHblits, CCMPred, ESMFold, AlphaFold2, and CLEAN, but does not provide specific version numbers for these software components or a comprehensive list of software dependencies with versions. |
| Experiment Setup | Yes | Training Hyperparameters We use the same Transformer (Vaswani, 2017) module as ESM-2 to implement Profile BFN. For the Profile BFN model with 650 million parameters, it has 33 layers of 20-head self-attention blocks. The hidden and embedding dimensions are 1280, and the feedforward hidden size is 5120. Note that, unlike the ESM-2 model, we do not use any form of dropout for regularization, as the Bayesian flow itself provides sufficient stochasticity. For the Bayesian flow, β(1) implies the uncertainty of the last step in the modeling procedure. Based on our empirical experience and cases in the original BFN paper (Graves et al., 2023), we found it could be approximately set according to the equation beta(1) K = constant (K is the vocab size). With this principle, we could directly obtain a good setting of β(1) following the previous empirical parameter in Graves et al. (2023) where K is different. We consider three different candidate schedule functions for β(t), linear, square and exponential, then we enumerate all three settings empirically over the small model (8M) and find linear works best in our task. We use Adam W (Loshchilov, 2017) to train our model, setting the learning rate at 0.0001, which linearly decays to a minimum of 4e-5. We adaptively set the batch size to approximately 2 million tokens. |