Data-efficient protein language models craft ultrastable carbon-capture enzymes beyond evolution
データ効率的なタンパク質言語モデルが進化を超えた超安定な炭素回収酵素を創出 (AI 翻訳)
Ravi Kumar Verma, Yaiza Serrano, Wisely Chua, Takumi Ho, Elaine Tiong, Yee Hwee Lim, Fong Tian Wong, Hao Fan
🤖 gxceed AI 要約
日本語
本研究では、炭酸脱水酵素（αCA）に特化したファミリー特異的タンパク質言語モデルを開発し、実験的検証と組み合わせることで、従来のタンパク質設計法では到達不可能な超安定かつ高活性な炭素回収酵素を創出した。設計された酵素は95℃で安定で、ベンチマーク酵素と比較して1.8倍の活性向上を示し、自然進化や祖先配列復元よりも優れた性能を達成した。このアプローチは他のタンパク質ファミリーにも一般化可能な設計指針を提供する。
English
This study develops a family-specific, fine-tuned protein language model for α-carbonic anhydrases (αCAs) and integrates experimental validation to engineer ultrastable carbon-capture enzymes. The lead variant remains stable at 95°C and shows 1.8-fold higher activity than the benchmark enzyme tdCA S82Y, outperforming all natural and ancestral counterparts. The work provides a generalizable blueprint for designing robust enzymes for industrial carbon capture.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本はCCUSをGXの重要施策として位置づけており、酵素を用いた低コスト・高効率なCO2回収技術は国内の炭素回収コスト削減に貢献する可能性がある。本研究のAI駆動型設計手法は、日本の化学・素材企業が進めるバイオものづくりとも親和性が高く、今後の産学連携の方向性を示唆する。
In the global GX context
Globally, this paper demonstrates the first use of fine-tuned protein language models to create carbon-capture enzymes with industrial-level stability, addressing a key bottleneck in biocatalytic CO2 capture. The approach is a significant advance for the CCUS community and offers a template for engineering other enzymes under operational stress, aligning with the push for scalable negative emissions technologies.
👥 読者別の含意
🔬研究者:Provides a novel method combining protein language models with wet-lab validation to explore unexplored sequence space for enzyme stability and activity.
🏢実務担当者:Highlights a promising biocatalyst for CO2 capture that could be integrated into industrial processes, potentially reducing energy and cost requirements.
🏛政策担当者:Supports the development of innovative carbon capture technologies, reinforcing the need for R&D investment in AI-driven biotechnologies for climate goals.
📄 Abstract（原文）
α-carbonic anhydrases (αCAs) underpin industrial carbon capture, yet natural enzymes are notoriously unstable under operational stresses. To overcome this limitation, a family-specific, fine-tuned protein language model (PLM) is developed and integrated with experimental validation to explore αCA sequence space beyond traditional protein design approaches. Although αCAs already operate near diffusion-limited rates (kcat ≈ 10 5 -10 6 s -1 ), the lead designed variant exhibited unprecedented robustness, remaining stable at 95 ºC and achieving a 1.8-fold increase in activity relative to the premier benchmark enzyme tdCA S82Y (stable up to 80 ºC; kcat ≈ 10 6 s -1 ). Despite sharing only 65.8% maximum sequence identity with the reference dataset, the engineered αCA enzyme outperforms all natural and ancestral counterparts. Computation-guided mutagenesis reveals distinct ligand-binding features absent in previously characterized αCAs, providing mechanistic insight into the enhanced performance. These findings demonstrate the utility of fine-tuned PLMs to access unexplored protein design space and establish a generalizable blueprint for engineering diverse protein families.
🔗 Provenance — このレコードを発見したソース
- openalex https://doi.org/10.26434/chemrxiv.15004194/v1first seen 2026-06-04 05:01:28
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。