bfn-confidence-general-proteins
收藏资源简介:
BFN Confidence General Proteins (V6) 是一个用于微调贝叶斯流网络(BFN)在通用单链蛋白质结构上置信度头的数据集,以AlphaFold2作为教师模型。该数据集是最大版本,通过合并V5版本(包含通用蛋白质和疾病蛋白质)和脑疾病蛋白质数据集(涵盖26个类别)并去除重复条目构建而成。数据集包含2,032个条目,其中训练集有1,625条(约2.8 GB),验证集有407条(约667 MB),总计约3.5 GB。子集包括脑疾病蛋白质(训练713条,验证170条)和内在无序蛋白质(IDP,训练32条,验证5条),在合并过程中移除了240个重复项。每个条目以LMDB格式存储,为pickled字典,包含以下字段:pdb_id(PDB标识符字符串)、sequence(氨基酸序列字符串)、batch(预处理的BFN兼容输入字典)、af2_plddt(每个残基的pLDDT分数列表,范围[0,1])、af2_iptm(ipTM分数浮点数)、af2_pae_matrix(PAE矩阵,为L×L的浮点数列表列表)。所有置信度分数均来自AlphaFold2的预测。数据集适用于蛋白质结构置信度预测任务,特别是用于训练或微调BFN模型的置信度头。数据源基于与ColabFold(AlphaFold2)预测匹配的PDB结构。
BFN Confidence General Proteins (V6) is a dataset for fine-tuning Bayesian Flow Networks (BFN) on confidence heads for general single-chain protein structures, using AlphaFold2 as the teacher model. V6 is the largest version of this dataset, constructed by merging V5 (which includes general proteins and disease proteins) and brain disease proteins (covering 26 categories) and removing duplicate entries. The dataset contains 2,032 entries, with a training set of 1,625 entries (approximately 2.8 GB) and a validation set of 407 entries (approximately 667 MB), totaling about 3.5 GB. Subsets include brain disease proteins (713 training, 170 validation) and intrinsically disordered proteins (IDP, 32 training, 5 validation), with 240 duplicates removed during merging. Each entry is stored in LMDB format as a pickled dictionary, containing the following fields: pdb_id (PDB identifier string), sequence (amino acid sequence string), batch (preprocessed BFN-compatible input dictionary), af2_plddt (list of pLDDT scores per residue, range [0,1]), af2_iptm (ipTM score as a float), and af2_pae_matrix (PAE matrix as a list of lists of floats, L×L). All confidence scores are derived from AlphaFold2 predictions. The dataset is suitable for protein structure confidence prediction tasks, particularly for training or fine-tuning confidence heads of BFN models. The data source is based on PDB structures matched with ColabFold (AlphaFold2) predictions.




