Finalist / shortlisted
Original award wording
Finalist
Project profile
Extremolith
A protein-research data and model pipeline with a web demo that returns a fixed, precomputed example.
The creator says Claude helped build a workflow for studying adaptation to extreme environments. The repository describes data integration, checks on environmental labels and controls for phylogenetic bias. We infer that the project is aimed at researchers; this has not been confirmed through user research.
According to the README, the web demo returns precomputed laccase results regardless of the sequence entered. We did not run model training, inference or a biological experiment. The current repository may also differ from the competition submission.
Creator or team (original public names)
Jaymin Patel (event credit); GitHub: jayman1466.
Links & sources
Links may have changed since we checked them. Descriptions from creators have not been independently tested.
Details in the original language
Original title
Extremolith
Original creator credit
Jaymin Patel（赛事公开署名；关联 GitHub 账号 jayman1466）
作品是什么
Extremolith 是蛋白质研究软件，作者在赛事作品页介绍用 Claude 辅助构建数据处理、模型训练与生成流程。仓库 README 与既有作品链接互相对应。
给谁用
从研究问题和输入输出理解，适合研究酶与极端环境适应性的研究人员；这是编辑归纳，并非已验证的用户画像。
问题或体验
作者希望研究酶在极端温度、酸碱度或盐度下的适应性；功能保留与泛化效果仍是作者陈述，没有独立实验验证。
如何使用
README 列出从 GTDB 整合蛋白质与环境信息，结合 GenomeSPOT 预测、分泌蛋白筛选及训练数据划分的命令行流程；赛事介绍另述模型训练和生成步骤。
得到什么
仓库列出带环境标签的数据表与蛋白质序列文件；网页演示仅返回木质素降解漆酶的预计算结果，不随输入序列实时生成。
关键机制
作者说明用元数据与预测结果交叉核对环境标签，并通过亲缘配对和训练集划分控制混杂与泄漏；这属于公开设计说明，未运行代码验证。
特点
将环境标签可信度与亲缘偏差控制纳入数据准备。公开网页演示不是实时模型服务，当前仓库也不能直接代表比赛提交时的版本。
Events & award announcements
Read original announcement
Sources last checked: 2026-10-21