Bionet applies AI agents and Harness engineering in biotech R&D
An iThome report says Bionet has integrated AI Coding and agents into bioinformatics workflows for exosome, drug and peptide research. Its Harness engineering framework governs tool access, data quality, validation, and software and data versions to support traceable and reproducible analysis. The report says Bionet published the DeepExoMir model in July to score and rank potential miRNA target sites for laboratory testing. Bionet also says it plans to develop a biological world model that predicts biological responses; the report does not describe a completed model or independently evaluated results.
The report presents a concrete framework for governing AI-assisted life-science analysis, including agent-assisted bioinformatics, peptide candidate narrowing, traceable and reproducible workflows, and RNA target-site scoring and ranking. The performance and impact of these applications are largely company-reported.
COVID-19疫情期間,清冠一號從研發到問世只花了5個月。國家研發團隊不只搶快,也想知道為何這帖中草藥複方對COVID-19有效。
這不是看單一藥材或查幾篇論文就能解決的難題。清冠一號包含多種中草藥成分,COVID-19牽涉病毒蛋白、人體蛋白質標的、免疫反應和發炎路徑,若用傳統方式,研究員得逐篇閱讀文獻、拆解複方成分,再一一比對可能的作用機制。
很多人不知道,訊聯生技集團當時也參與了這項療效分析研究,運用AI輔助生物資訊分析,在24小時內完成中草藥成分與逾3萬個人體及COVID-19相關蛋白質標的比對,協助找出清冠一號可能的作用路徑,相關成果後續發表在《Biomedicine》等期刊。
這不是訊聯第一次把AI用在生物資訊分析。早在2010年,訊聯就開始布局AI與生物資訊分析,自行開發機器學習演算法的分析程式,來處理海量的生物數據。
生成式AI興起後,他們進一步導入AI Coding和Agent,並用Harness工程控管工具、資料和版本,確保AI產出品質,將這套做法應用於外泌體、新藥與胜肽研發流程。訊聯將這套可追蹤、可驗證、可迭代的AI輔助方法融入生科研發,就像是生物資訊分析的AI Coding,並稱之為Bio-coding。接下來,他們要用這套研發模式,打造出能預測生物反應的「生物世界模型」。