PepPrCLIP 数据集

该数据为论文「De novo design of peptide binders to conformationally diverse targets with contrastive language modeling」（已于 2025 年 1 月发布于 Science Advances）所有原始和处理后的数据。该论文杜克大学的研究团队发布的成果，构建了基于 CLIP 的肽优先级筛选流程 PepPrCLIP，可以设计短蛋白质以结合和破坏以前无法用药的致病蛋白质。与使用目标 3D 结构生成肽的现有平台 RFDiffusion 相比，PepPrCLIP 速度更快，并且能够创建几乎总是与目标蛋白质更匹配的肽。

This dataset contains all raw and processed data from the paper titled "De novo design of peptide binders to conformationally diverse targets with contrastive language modeling", which was published in *Science Advances* in January 2025. This work was developed by a research team from Duke University, where they constructed a CLIP-based peptide prioritization screening pipeline named PepPrCLIP. This pipeline enables the de novo design of short proteins to bind and disrupt previously undruggable pathogenic proteins. Compared with the existing peptide generation platform RFDiffusion which relies on target 3D structures, PepPrCLIP features faster inference speed and can generate peptides that almost always exhibit better compatibility with target proteins.

搜集汇总

背景与挑战

背景概述

PepPrCLIP数据集源自杜克大学研究团队在Science Advances上发表的论文，包含用于设计短蛋白质结合致病蛋白的原始和处理数据。该数据集基于CLIP技术构建，相比现有方法速度更快且匹配度更高，大小为4.5 GB，采用CC BY 4.0许可证，属于AI for Science领域。