10x Genomics’ Post

🔇 The latest from EurekAlert! Insilico open sources #Precious3GPT, a novel multimodal multi species multi #omics multi tissue transformer for #agingresearch and drug discovery Precious3GPT is our most capable model for aging research so far. Unlike previous Precious-series models, Precious3GPT is a collaborative effort. We collaborated with Vadim Gladyshev and Albert Ying at Harvard to develop the concept and are making the model available for the community via Hugging Face and GitHub with the pre-print describing the model and its initial capabilities. As the community starts using the model, we are planning to improve it and add additional features and capabilities. Read the full release here: https://lnkd.in/dVtgFTfH

In case-control single-cell transcriptomic experiments, not all of the cells from a given case will be affected by the experimental perturbation, reducing the accuracy of standard analysis methods. Aleksandrina Goeva, Evan Macosko, and colleagues designed HiDDEN, a computational method that refines case-control labels to more accurately reflect each cell’s perturbation status and detect subtle perturbation effects in single-cell data. The team demonstrated HiDDEN’s capabilities on datasets from human multiple myeloma precursor conditions and a mouse model of demyelination. Read more about HiDDEN and its potential applications in Nature Communications. #BroadInstitute #Science #ScienceNews #Research #ScientificResearch

I love this idea of using machine learning to derive insights from data which may otherwise be considered noise, as opposed to falling into the common pitfall of assuming important data to be outliers and throwing out important information.

"bioinfoRmatics v2" with Budapest Users of R Network Date: Tuesday, April 16, 2024 6:00 PM -9:00 PM CEST Location: CEU Nador 15 building, Nádor utca 15, Budapest https://lnkd.in/eXBrdsRP Details Based on the very positive feedback on the previous R meetup, we are hosting another session dedicated to bioinformatics and life sciences, and I am very much looking forward to meeting you all on April 16! Although we already have 5 lightning talks scheduled, if you are interested in giving a 5-15 mins talk (freestyle, either in Hungarian or English), please let us know. If you are not willing to speak this time -- no worries; please join us for the talks and the networking event afterward. Schedule: 18:00 doors open 18:15 talks @ N15 101 (1st floor) 19:30 networking (pizza, soft drinks) @ N13 Panorama Lounge (7th floor) 20:30 end of meetup, optionally walking to a nearby pub List of planned talks: Kristof Kelemen: Hungarian Forest Reserves Shiny app We wanted to make point-based data from forest reserves publicly available. I will talk about the difficulties of presenting multivariate spatial data and discuss further plans to make the site better. The app can be visited at https://lnkd.in/eKcxBP-c Tibor Nagy (MATE): When NOT to use R R is a good language for many purposes, but sometimes we need to use something else. In this presentation, I will show use cases when not using R is more efficient. Tamás Hegedűs (Semmelweis): 3D-bioinformatics in the AlphaFold era DeepMind's AlphaFold revolutionized the field by accurately predicting protein structures from sequences, solving a decades-old challenge. This breakthrough has not only made high-quality structures of all known proteins, but also significantly advanced protein design and drug development. The impactful journey of DeepMind tools will be encapsulated through relatable, everyday examples (e.g. https://lnkd.in/eZ7aAjHC). Teadora Tyler (Semmelweis): CellChat: predicting cell-to-cell communication in single cell RNAseq data The data generated by single cell RNA sequencing in a gold mine. Recently, several packages were developed for predicting cell-to-cell communication pathways based on large databases of known ligand-receptor pairs. CellChat is a user-friendly package with innovative graphic solutions compatible with widely used single cell pipelines. Gergely Daroczi (Rx Studio): LLM for dosing suggestions? Rx Studio's background and the main precision dosing webapp for clinicians was briefly presented at a previous R meetup, but with the recent AI/LLM hype and related investor expectations, we piloted a patient-facing chat application for assisting paracetamol dosing. This talk will share some of the related concerns and experiences, and how we managed to build a chat MVP after heavily regulating LLM functionalities. #rstats #datascience #bioinformatics

This is interesting > The protocol repurposes an Illumina NovaSeq 6000 flow cell for spatial transcriptomics, generating high-resolution datasets and integrating a streamlined data-analysis pipeline.

We'll be sharing more about Nosis over the coming months with a number of big announcements, but in the meantime, I'm very excited to share an important contribution from our machine learning team to the field of drug design, Deep Kernel Inversion (DKI). For context, one of the key challenges in the development of new precision medicines has been a toolset for designing targeted and specific therapies that act only where they are intended and avoid off-target negative side effects. With tools like Alphafold, it is computationally infeasible to calculate targeted specificity, but with DKI, we can now reliably design highly targeted and specific therapies with exceptional on-target engagement. More broadly, we see potential applications for DKI in novel target discovery, the modeling of biological networks, and even modeling entire virtual cells and organs. We're still in the early stages of innovation in this field and looking forward to collaborating with others on continuing to advance methods for drug design! A big thank you to my co-founder Seth Myers for leading the effort in this work and building out this critical component of our approach to precision medicines.

Thrilled to share our latest work at Nosis Bio: Deep Kernel Inversion (DKI)—a new deep learning algorithm that makes analysis of interactions across proteomes 100,000 times more efficient than current methods. DKI represents a huge leap forward for the computational mapping of interactions networks, enabling the design of precisely targeted medicines. https://lnkd.in/gHW7Sf-r

An article published in BMC Bioinformatics presents ScRNAbox: a comprehensive end-to-end pipeline designed to streamline the processing and analysis of single cell RNAseq data.

What is the future of data imputation? In our new comment published in Nature Biomedical Engineering, we introduce a new framework, synthetic multimodal modelling. Foundation models are leveraged to model a patient status grounded in real data and complemented with synthetic data from those non-available modalities. Our framework rests upon four core principles:1. Joint embedding space and N to N generation: Like ImageBind, we envision our framework leveraging a shared embedding space, flexibly enabling the generation of any modality from any other.2. In-silico hypothesis testing: This joint embedding space allows conducting in-silico experiments, perturbing data to test scientific hypotheses about a patient's condition.3. Evolving patient data:Our framework accommodates the continuous integration of new patient data through time, thereby allowing the modeling of disease progression.4. Disease understanding:The joint embedding space allows researchers to unravel the complex relationships between diverse modalities, fostering a deeper understanding of the disease itself. You can find it here: https://lnkd.in/dPYVPgq6

RhoFold+ sounds like a groundbreaking development in RNA research. By accurately predicting the 3D structures of single-chain RNAs directly from sequences, it has the potential to greatly enhance our understanding of RNA folding, interactions, and function. A must-read for all RNA enthusiasts!