kotaro_chato

統計/データサイエンス/化学/マテリアルズインフォマティクス

[Ryousuke_Wayama @wayama_ryousuke]

https://t.co/h5dZSDkr6S

[ハヤカワ五味 @hayakawagomi]

去年からずっと「生成AIをガチることは(経営レベルの)意思決定量が100倍になること」と言っていたけど、それを分かりやすく言語化されていて良かった。

何より「面倒さ＝通行料」という整理がすごい分かりやすい。 https://t.co/gKCgMHOzkV

[Osamu OGASAHARA @ogasahara]

「さくらのAI」の第一弾となるパブリックサービス　「さくらのAI Engine」を出しました。
無償枠もあるのでぜひお試しを〜。
OpenAI準拠のChat completions APIやWhisperのAudio transcription APIとかとか。
モデルはQwen3-Coder-480Bとかもありますよ。

使ってみて、プライベートGPU環境でつかいたいとか、これらの上にサービス構築したいとかあれば、なんなりとご相談ください！
そしてなによりもフィードバックください！
使いたくなるサービスをつくり続けたいメンバーが待ってます！

[moritalous | Kazuaki Morita @moritalous]

記事を投稿しました！ Docker MCP Gatewayがすんばらしい👌 on #Qiita https://t.co/v70fRNsilZ

[Yehlin Cho @ChoYehlin]

🚀 Excited to release BoltzDesign1!

✨ Now with LogMD-based trajectory visualization.
🔗 Demo: https://t.co/tjBXSibZhT
Feedback & collabs welcome! 🙌

🔗: GitHub: https://t.co/HsQBrB8amQ
🔗: Colab: https://t.co/TzX4q5m2qp

@sokrypton @MartinPacesa

[Biology+AI Daily @BiologyAIDaily]

SCAFFOLDGPT: A Scaffold-based GPT Model for Drug Optimization

1. SCAFFOLDGPT is a new GPT-based framework designed to optimize drugs by preserving the core scaffold of existing molecules while improving pharmaceutical properties like druglikeness, synthesizability, docking scores, and solubility.

2. Unlike De Novo generation, SCAFFOLDGPT focuses on refining FDA-approved drugs, making it more efficient for fighting rapidly mutating viruses (like SARS-CoV-2) and drug-resistant cancers.

3. The core innovation lies in its three-stage pipeline: pretraining on SMILES data, reinforcement learning finetuning with Advantage-aligned Policy Optimization (APO), and a novel token-level decoding strategy called TOP-N.

4. The two-phase incremental pretraining scheme first trains the model on molecules alone, then on scaffold–molecule pairs, effectively teaching the model to generate molecules conditionally based on a given scaffold.

5. The second stage employs APO to align molecule generation with multiple objectives: druglikeness, solubility, synthesizability, docking score, and scaffold similarity, using a reward function built from ensemble critics.

6. The third stage, TOP-N decoding, improves controllability by guiding token selection based on reward optimization rather than just likelihood, outperforming traditional decoding methods like top-k or nucleus sampling in benchmark tests.

7. Compared to state-of-the-art methods such as REINVENT4 and DrugImprover, SCAFFOLDGPT achieves superior results on cancer and COVID datasets, balancing property enhancement with scaffold preservation.

8. Ablation studies show that both APO and TOP-N significantly contribute to performance gains, while the incremental training improves generation validity—especially for longer sequences.

9. SCAFFOLDGPT demonstrates consistent property improvements across benchmarks while retaining high similarity to original molecules, making it a practical and effective tool for real-world drug optimization.

💻Code: https://t.co/xEoadoh5ac
📜Paper: https://t.co/bbx92E3ZQT

#DrugDiscovery #AI4Science #MolecularDesign #GenerativeAI #ReinforcementLearning #GPT #ComputationalBiology #DrugOptimization

[横山トモヤス｜計算材料科学者 @yoko_materialDX]

MI-6さんのメディアmiLabにて「材料科学における夢の技術、結晶構造予測の現在地」というタイトルで記事を執筆させていただきました。

マテリアルズインフォマティクスの分野で最近流行ってきている結晶構造予測の面白さや難しさが伝われば嬉しいです。
https://t.co/yArsGLkDVS

[島らっきょう大好きバナナさん🍌 @SimarakyoBanana]

✅GPCRバイアスアゴニストとしてのリガンド-ナノボディ融合体

クラスB GPCRはリガンドと2ヶ所で結合してGタンパク質とβ-アレスチンの経路を両方とも活性化します🔥

そうすると薬効に関わる経路と副作用に関わる経路が同時に活性化されてしまうので、薬効に関わるほうだけを活性化するようにしたのがバイアスアゴニストです！

この論文では、骨粗鬆症の治療標的であるPTHR1（副甲状腺ホルモン受容体1）を題材に検証してます💁‍♂️

まずPTHR1リガンドについて整理↓
　・PTH1-34（全長）:サイト1/2結合→ 高親和性だが骨形成と骨吸収の両方を誘導
　・PTH1-11（部分構造）: サイト2結合→低親和性だが骨形成を誘導

そこで、サイト1に結合するアゴニスト活性のないナノボディとPTH1-11を融合することで高親和性かつサイト2だけに結合して副作用を減らして骨形成を誘導することができる訳です👏

この方法はGLP1R, CRHR1などの似た構造のGPCRにも応用できる戦略なので良いですね〜👀

[drug hunter @drughunter_com]

Icotrokinra is a potential first-in-class, orally bioavailable macrocyclic peptide antagonist of IL-23R, currently in clinical trials for multiple inflammatory diseases.

This article highlights compelling late-stage clinical trial data supporting icotrokinra’s potential to rival biologics, and explores how this molecule could reshape treatment paradigms for a range of inflammatory conditions.

Read more: https://t.co/W3thL7jSzi

[Arne Schneuing @rneschneuing]

The code & camera-ready version of our #ICLR2025 paper on "Multi-domain Distribution Learning for De Novo Drug Design" are now available

📚 Paper: https://t.co/NCeHVltqm8

💻 Code: https://t.co/0GGMiOODla

(1/4)

[吉村洋文（大阪府知事） @hiroyoshimura]

出生数、初の70万人割れへ　24年、1～11月は66万人
→0歳児選挙権やってみません？政治家が、口だけでなく、本気で次世代、子育て世代の方を向くようになると思います。 https://t.co/kVhJnMMGow

[Biology+AI Daily @BiologyAIDaily]

Free online course:
MLCB24 - Machine Learning in Computational Biology Fall 2024 from @manoliskellis @MIT

https://t.co/O4PsH0X4fh

L1: Course Overview, Machine Learning, Deep Learning, Generative AI, and the Unification of Biology

Module1: Genomics，Epigenomics，Single-Cell，Networks，Circuitry
L2: Expression Analysis, Clustering/Classification, Gaussian Mixture Models, K-means, Bayesian Inf, Gen-vs-DiscrML
L3: Sequential Data, Alignment, DynProg, Hidden Markov Models, Parsing, Posterior Decoding, HMM architectures
L4: Regulatory Genomics: Motifs, Information, ChIP, Gibbs Sampling, EM, CNNs for Genome Parsing
L5: Epigenomics: Peak calling, Chromatin states, 3D structure, Hi-C, Genome Topology
L6: Single-cell genomics, sc-multi-omics, non-linear embeddings, spatial transcriptomics, next-gen technologies
L7: Regulatory Networks: Graphs, Linear Algebra, PCA, SVD, Dimensionality Reduction, TF-enhancer-gene circuitry

Module2: Protein Structure, Protein Language Models, Geometric Deep Learning
L8:Intro to Structural biology
L9: Protein structure and folding:Diffusion models,Cry-EM,Protein design
L10: Intro to transformers and Large Language Models LLMs
L11: Protein Language Models PLMs and Transfer Learning
L12: DNA language models:Chromatin Structure

Module3: Chemistry, Therapeutics, Graph Neural Networks
L13: Overview of drug development
L14: Intro to small molecules
L15: Representation of small molecules:Graphs,GNNs,Transformers,RDKit
L16:Docking: Small molecule-proteins docking
L17:Disease Association Mapping, genetics, GWAS, linkage analysis, disease circuitry, variant-to-function
L18:Quantitative trait mapping, molecular traits, eQTLs, mediation analysis, iMWAS, multi-modal QTLs

Module4: Electronic Health Records, Imaging,Evolution, Metabolism
L19:Electronic Health Records, AlIOFUs, UKBioBank, Medical Genomics, Pop-Scale Cohorts, Multi-Ancestry
L21:Imaging methods for biological applications
L22:Comparative genomics, Conservation, Evolutionary signatures, PhyloCSF, RNA structure, Motif BLS2conf
L23:Gene families, Homologous, Duplication, RNA world, RNA folding, lincRNAs, RNA modifications, m6A
L24:Modeling metabolism: Flux balance analysis
L25:Measuring metabolism: Metabolomics and Deep Learning

[K.Oki@製造系データサイエンティスト @oki_kosuke]

「メタ、材料科学向けの最大規模のデータセットとAIモデルを無償公開」
1億1000万のデータポイントが含まれ、材料科学分野の研究加速が期待されるとのこと。
グーグルやマイクロソフトはモデルのみの公開だが、メタはこの公開でどのような展開を目論んでいるのだろうか？
https://t.co/J3TyJIResn

[Biology+AI Daily @BiologyAIDaily]

A long context RNA foundation model for predicting transcriptome architecture

• LoRNASH redefines RNA modeling: This foundation model predicts transcriptome architecture directly from long RNA sequences, capturing both structure and abundance of RNA isoforms with unparalleled accuracy.

• Groundbreaking long-context capability: LoRNASH processes sequences up to 65,000 nucleotides, enabling precise modeling of complex RNA interactions, overcoming limitations faced by previous models like SpliceAI and Nucleotide Transformer.

• Leveraging long-read sequencing data: The model is trained on PacBio IsoSeq data from 26 cancer cell lines and incorporates both human and mouse datasets, totaling 7 billion tokens across 297,726 unique transcripts.

• Zero-shot performance excellence: LoRNASH achieves remarkable success in diverse prediction tasks without fine-tuning, including absolute isoform abundance, exon trapping, and pathogenic non-coding variant analysis.

• Next-level generative capability: LoRNASH can generate synthetic transcripts that mirror real-world RNA structures, opening avenues for transgene design and splicing regulation studies.

• Enhances biomedical research: With high precision in modeling alternative splicing and identifying cis-regulatory elements, the model accelerates our understanding of transcriptome dynamics and disease-associated variations.

@genophoria @hsnajafabadi @AminMEmad @vram142 @MNaghipourfar @ahcorcha @choi_benedict
💻Code: https://t.co/nHRhAHbJyD
📜Paper: https://t.co/Ynu7ZDclCX

[株式会社リクルート データ推進室 @Recruit_Data]

【ブログ記事公開のお知らせ】
kaggleのISIC 2024で5位、金メダルと賞金1万ドルを獲得しました
https://t.co/tNJZF4l8XW

記事では、コンペのタスクと解法について解説しています！
是非ご覧ください✨
#kaggle

[Ray Fernando @RayFernando1337]

@rohanpaul_ai I’ve made a calculator that does a similar estimate so you can change things in the UI.

App: https://t.co/AjKen1qFQg

Open Source code:
https://t.co/ZHSbb6OmoX

[TJO @TJO_datasci]

ブログ更新しました。 / 実務において回帰分析を行うに当たっての注意点を改めて挙げてみる - 渋谷駅前で働くデータサイエンティストのブログ https://t.co/KaAXF7suBz

[Google DeepMind @GoogleDeepMind]

Announcing AlphaFold 3: our state-of-the-art AI model for predicting the structure and interactions of all life’s molecules. 🧬

Here’s how we built it with @IsomorphicLabs and what it means for biology. 🧵 https://t.co/gjw6Ip4F2M

[LabCode（ラボコード） @LabCodeBlog]

【新記事公開！】
In silico創薬のLigand-based screeningについての記事を公開しました！

話題のプベルル酸の毒性予測とインフルエンザ治療薬への可能性を模索しています。

ぜひ試してみてください！

https://t.co/2vyXZV9Bwn

#LabCode #Insilico創薬 #QSAR #LigandBasedscreening