Dr. SoyBiotec 👨🏻‍🎓🥳

1 like ❤️ = 1 concepto de genética 🧬 🔽🔽

#NextGenSci asked young scientists to write a rule that all principal investigators should be required to follow to improve the experience of young scientists in their lab. Check out their answers on #NationalHigherEducationDay: scim.ag/3HoP39O

Why are there intronic reads in your bulk RNA-seq data? You're not alone—it's common, and the reasons are more layered than you think. Let’s break it down. 🧵

@ATinyGreenCell So...when are you making one? 😉😉

Dear god lol

TrialMatchAI: An End-to-End AI-powered Clinical Trial Recommendation System to Streamline Patient-to-Trial Matching １．TrialMatchAI is a modular, fully open-source clinical trial recommendation system designed for high-accuracy, privacy-preserving patient-trial matching using fine-tuned LLMs and a retrieval-augmented generation (RAG) framework. ２．Unlike black-box GPT-based systems, TrialMatchAI runs locally, complies with GDPR and HIPAA, and integrates seamlessly with hospital EHRs via the Phenopackets data format, enabling secure real-world deployment. ３．The system processes both structured data (age, diagnosis, labs) and unstructured text (physician notes, molecular reports) to extract and normalize biomedical entities for precise trial matching. ４．It uses a hybrid search strategy: BM25 for keyword-based retrieval and dense vector search with BGE-M3 embeddings to capture deeper semantic relationships in eligibility criteria. ５．Two fine-tuned LLMs—Gemma-2-2B and Phi-4—handle trial re-ranking and eligibility classification using medical Chain-of-Thought reasoning, providing interpretable, criterion-level justifications for inclusion/exclusion decisions. ６．Evaluation on TREC 2021 and 2022 datasets (over 26,000 trials each) shows TrialMatchAI retrieves >90% of relevant trials in the top 3% of results and achieves top-nDCG and precision@k scores, outperforming prior state-of-the-art systems. ７．In a synthetic “Ideal Candidates” benchmark, 95% of patients had their ground-truth trial ranked in the top 2; in real-world evaluation on 52 cancer patients from the NKI WIDE study, 92% had at least one matching trial in the top 20. ８．Criterion-level accuracy exceeds 90% across inclusion and exclusion criteria, with the model particularly excelling in biomarker-driven trial matching—validated on molecular profiles extracted from MTB reports. ９．TrialMatchAI’s modular design supports plug-and-play replacement of LLM components, entity extractors, and retrievers, ensuring adaptability as biomedical LLMs evolve. １０．The system’s outputs include JSON-structured explanations for each criterion, allowing human oversight and enabling trust in AI-driven decisions—key for adoption in clinical oncology workflows. １１．TrialMatchAI demonstrates that small, fine-tuned, open-source models can match or exceed closed LLMs in clinical NLP tasks, and sets a new bar for explainable, scalable AI in trial recruitment. @MachaNikolski @headinthebox @rjafijneman @rdienstmann @LanaSMeiqari @KarkarSlim @agroppi @sigven @majdabd33 💻Code: github.com/cbib/TrialMatc… 📜Paper: arxiv.org/abs/2505.08508 #ClinicalTrials #AI4Science #LLM #BiomedicalNLP #PatientRecruitment #TrialMatching #Bioinformatics #PrivacyPreservingAI #RAG #PrecisionOncology

New preprint from our lab! TrialMatchAI an AI-driven patient-to-trial matching system, with outstanding work by @majdabd33. It uses fine-tuned LLMs + medical chain-of-thought reasoning. #LLM #ClinicalTrials #PrecisionMedicine #TrialMatchAI #Bioinformatics arxiv.org/abs/2505.08508…

@ATinyGreenCell Thanks! Makes sense...🤔

@SoyBiotec 15 months on average if mechanism of action is not some crazy crap. Precedence is king so often folks just inch ahead by using well studied parts from well studied organisms. Its still very hard to deregulate truly novel traits.

Here is a list of all the cool plant bioengineering products being actively deregulated by the USDA, on the path to commercial approval. Enjoyyyy. aphis.usda.gov/biotechnology/…

@BreakingVlad Un abrazo muy grande, Vlad

De los regalos más increíbles que hemos visto en el programa. Espero que no acabe en el camerino de Broncano junto a la camiseta de Pito Camacho. #LaRevuelta @JaneGoodallEspa

@SynBio1 In Spanish it's funnier than in English 😂

Pedobacter needs to fire their PR team and rebrand

Brian Greene on Cutting Funding for Science and Innovation.

Damos la enhorabuena a Cristina Diaz Gonzalez por haber realizado la mejor presentación del pósters dentro del Máster de Biología y Clínica del Cáncer del curso 24-25

La transcriptómica (Disciplina que estudia el ARNm de las células) ha evolucionado en los últimos años, tanto, ¡que ahora hay 3 enfoques principales para planificar tus experimentos! 🧬💻 #Transciptómica #Bioinforma2 #RSGSpain #Bioinformatica #BiologiaComputacional #Divulgación

Super excited to share my first PhD preprint! Meet LySE - a novel phage-based system for directed evolution of entire gene clusters in E. coli, overcoming size limitations and off-target mutations in classical directed evolution 🧬 👉🏻 biorxiv.org/content/10.110… 1/

Como profesor de la Facultad de Físicas de la @unicomplutense, hoy participé en una acción simbólica contra los recortes del gobierno de la Comunidad de Madrid: desplegamos una pancarta con el mensaje “Universidad pública, inversión en calidad”. No fue una pancarta cualquiera…

@ATinyGreenCell I'm thinking oligos for mushroom genotyping 👀

I'm placing a large oligo order to expand my parts library both for clients and myself. Any requests? Gonna add my melanin gene, some lux operon bits for glowy microbes, and a few petunia homeotic genes. Fun ideas you'd like to see me try or offer in my shop? Lemme know!

this is actually incredible

@oscar_puente_ @elperiodico ¿Los científicos españoles en el extranjero cuentan?

España destinará 45 millones de euros para atraer a científicos extranjeros en pleno huracán Trump elperiodico.com/es/ciencia/202… a través de @elperiodico

Mal día para los banana breads