Genki Terashi

Very excited to share our new paper DAQ-score, just published in @naturemethods . Our deep learning method can estimate the local quality of the protein structure model from cryo-EM data. DAQ is available on Google Colab: bit.ly/daq-score Paper: lnkd.in/eTvJfSvH

We have a postdoc opening. Please check careers.iscb.org/jobs/view/9910

Our new book chapter has just been published! In this chapter, we introduce our recent work in computational structural biology and practical applications of our tools available on em.kihara.org. You can read the chapter here: authors.elsevier.com/a/1ms5k6gpflQm…

We’ve rebuilt our Protein Function Prediction server! New: • Domain-PFP: self-supervised, domain-aware function prediction • GO2Sum: converts GO terms into UniProt-style functional summaries All methods (PFP, Phylo-PFP, ESG) now in one interface. kiharalab.org/pfp

Intellicule is hiring a part-time Software Engineer to advance our computational biology tools for protein design, nanobody engineering, and Cryo electron microscopy (cryo-EM) based drug discovery. Submit resume and cover letter to contact@intellicule.com

New collaboration paper with the Bou-Abdallah lab at SUNY Potsdam: "Ferritin iron uptake and oxidation are dynamically modulated by nucleotide phosphate architecture via electrostatic gating", International Journal of Biological Macromolecules. sciencedirect.com/science/articl…

Accurate Macromolecular Complex Modeling for Cryo-EM with CryoZeta biorxiv.org/content/10.648… #biorxiv_bioinfo

🚀 First DAQ-Score DB update of 2026 is live! Cryo-EM protein model quality evaluations now cover 266,577 protein chains. 🔗 daqdb.kiharalab.org 🏁Check out also our New ChimeraX plugin! It lets you check DAQ scores during modeling! cxtoolshed.rbvi.ucsf.edu/apps/chimeraxd…

DAQ-Score、新春アップデート第一弾🎍 ChimeraXプラグインも作りました！ モデリングしながらDAQ-Scoreをinteractiveに確認できます。ぜひ論文で構造バリデーションにご利用ください。

DAQ for ChimeraX is now released! DAQplugin computes residue-wise DAQ scores directly in ChimeraX. By comparing cryo-EM maps with atomic models, users can quickly identify potential problems, such as sequence shifts or incorrectly assigned residues. cxtoolshed.rbvi.ucsf.edu/apps/chimeraxd…

December Update of DAQ-Score DB! Quality evaluations of protein models from cryo-EM. Now includes evaluation of 258,956 protein chains. Check at daqdb.kiharalab.org It is very easy to compute DAQ for your protein model: em.kiharalab.org/algorithm/daqs…

New paper in collaboration with Jinghui Luo lab at PSI Paul Scherrer Inst. "Structural Insights and Functional Dynamics of β-Lactoglobulin Fibrils", @NanoLetters We used DeepMainmast for structure modeling. pubs.acs.org/doi/full/10.10…

🔬 *Advancing Precision Medicine* 🗞️ Congrats to Purdue Institute for Cancer Research member Daisuke Kihara and his team at Intellicule on receiving a Small Business Innovation Research Phase I grant from the @NIH. The project aims to expand and advance structural modeling and analysis for drug discovery using cryo-EM by utilizing state-of-the-art deep-learning techniques, helping scientists design better, more targeted treatments for cancer and other diseases. purdue.link/intellicule-st…

Intellicule, which licenses @LifeAtPurdue intellectual property, has received a $217K SBIR grant from the @NIH to develop software that could impact precision medicine. Read more about it: bit.ly/4aBmCSg #Purdue #PurdueInnovates

Intellicule receives NIH grant to develop biomolecular modeling software - Purdue University - Kihara is a professor of biological sciences and computer science in Purdue University's College of Science. He also is a member of the Purdue ... - ift.tt/JtmM0cF

博士号取得を目指す50歳以上の方に向けた支援事業です。国籍、性別、文系理系を問わないそうです。締切は12月5日です。 gllc.or.jp/project/doctor…

【#UJA論文賞2026 応募スタート！】 海外で研究に挑むすべての若手研究者へ── 医学医療系も、化学も、物理も、生物学も。 工学も、情報科学も、人文社会科学も。 北米でも、欧州でも、アジアでも。 どの分野で、どこにいても。 あなたの研究には、価値がある。 あなたの挑戦を、私たちは応援したい。 過去の受賞者たちも、一本の論文から新しい扉を開きました。 📌 応募対象：過去2年間に発表した留学先での筆頭著者論文 🌏 対象：全分野・全地域 🏆 特典：表彰, 研究発表の機会, トラベルグラント 応募締切：2025年11月30日 cheironinitiative.wixsite.com/uja-award 海外日本人研究者ネットワーク（UJA）は、 あなたの挑戦を、心から応援しています。 #UJA論文賞 #海外研究 #若手研究者 #研究留学 #全分野募集 #ポスドク #博士課程 #PhDstudent #Postdoc

New paper released! "Distance-AF improves predicted protein structure models by AlphaFold2 with user-specified distance constraints" Yuanyuan Zhang, Zicong Zhang, Y Kagaya, G Terashi, B Zhao, Y Xiong & D Kihara, Communications Biology. @CommsBio nature.com/articles/s4200…

新しい論文のリリースです。アミノ酸残基の距離情報を考慮してAlphafoldの構造モデルを修正するDistance-AF. 新しく距離制約を学習したネットワークを作るのではなく、距離制約のロスの項を足してオーバーフィッティングを行います。Communications Biology.

Thank you for coming to the DAQ and DAQ-Refine cryo-EM structure validation workshop hosted by @PurdueSBBC ! For those who missed it, we have Youtube video: youtube.com/watch?v=YRAT98… The presented tools are available at em.kiharalab.org

AlphaFold model quality self-assessment improvement via deep graph learning 1. Introducing EQAFold, an enhanced framework that refines the Local Distance Difference Test (LDDT) prediction head of AlphaFold to generate more accurate self-confidence scores for protein structure predictions. 2. The core innovation of EQAFold lies in its use of Equivariant Graph Neural Networks (EGNNs) to replace the standard LDDT prediction head in AlphaFold. This allows for better utilization of spatial and relational information within the protein structure graph, leading to improved confidence score accuracy. 3. EQAFold outperforms the standard AlphaFold in providing more reliable confidence metrics, especially in regions with substantial LDDT prediction errors. This is crucial for downstream applications as it helps researchers better gauge the reliability of predicted protein structures. 4. The training dataset for EQAFold was carefully curated to exclude entries where the polypeptide chain was extracted from a larger multimeric structure, ensuring more accurate evaluation of monomeric proteins. Additional features like fluctuations from multiple runs and embedding data from protein language models were also incorporated. 5. In benchmarking against the standard AlphaFold architecture on a test dataset of 726 monomeric protein structure entries, EQAFold demonstrated superior performance in both model-level and residue-level pLDDT assignments, with statistically significant improvements in categorizing residues with different true LDDT values. 6. EQAFold was also compared with another recent method, EnQA, and showed substantial improvements in LDDT error for the majority of cases. Even when retrained to predict the more stringent LDDT-AA metric, EQAFold-AA outperformed EnQA on a significant number of targets. 7. The authors suggest that similar approaches could be broadly applicable to various deep learning-based protein structure prediction tasks, including improved accuracy assessment for antibody prediction, protein complexes, peptide-docking, nucleic acids, and small molecules. 💻Code: github.com/kiharalab/EQAF… 📜Paper: onlinelibrary.wiley.com/doi/10.1002/pr… #AlphaFold #DeepLearning #ProteinStructurePrediction #EQAFold #ModelQualityAssessment