Allotrope Foundation’s Post

I am thrilled to announce that my latest paper has been accepted for publication! Titled "𝐂𝐨𝐧𝐬𝐞𝐧𝐬𝐮𝐬 𝐇𝐨𝐥𝐢𝐬𝐭𝐢𝐜 𝐕𝐢𝐫𝐭𝐮𝐚𝐥 𝐒𝐜𝐫𝐞𝐞𝐧𝐢𝐧𝐠 𝐟𝐨𝐫 𝐃𝐫𝐮𝐠 𝐃𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲: 𝐀 𝐍𝐨𝐯𝐞𝐥 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐌𝐨𝐝𝐞𝐥 𝐀𝐩𝐩𝐫𝐨𝐚𝐜𝐡" it will appear in the prestigious Journal of Cheminformatics. The paper reveals three main findings: 1. 𝑾_𝒏𝒆𝒘 𝐌𝐞𝐭𝐫𝐢𝐜: We introduce a mathematical formula that provides a new metric for machine learning models, "W_new." A high value of W_new indicates high performance metrics (R² for validation and training), low error metrics, and a reduced chance of overfitting. 2. 𝐂𝐨𝐧𝐬𝐞𝐧𝐬𝐮𝐬 𝐕𝐢𝐫𝐭𝐮𝐚𝐥 𝐒𝐜𝐫𝐞𝐞𝐧𝐢𝐧𝐠: We developed a workflow that combines four screening methods to identify the best binders in the screening pool, enhancing the accuracy and efficiency of virtual screening. 3. 𝐁𝐢𝐚𝐬 𝐀𝐬𝐬𝐞𝐬𝐬𝐦𝐞𝐧𝐭 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰: We propose a method to assess and quantify the bias between active and decoy compounds, ensuring the robustness of ML models against the vulnerabilities of traditional models. This was validated using benchmark datasets such as MUV. The workflow that refines the best ML model using W_new can be applied not only in drug discovery pipelines but also across all domains utilizing ML models. Read the open-access paper, enjoy it, and 𝐝𝐨𝐧'𝐭 𝐟𝐨𝐫𝐠𝐞𝐭 𝐭𝐨 𝐜𝐢𝐭𝐞 𝐮𝐬! The codes are available on the GitHub link: https://lnkd.in/gB_ZWEVv https://lnkd.in/gzqcWKf8? utm_source=rct_congratemailt&utm_medium=email&utm_campaign=oa_20240529&utm_content=10.1186/s13321-024-00855-8

Machine learning is a life saver in analysing and classifying mammoth data in medical science filed.

The word "database" appears NINETY-EIGHT times in the Tempus AI S-1. How many times does it appear in GRAIL's Form 10/S-1? ZERO. OMG they just don't get it. It's the data 🤯 I've long promoted the idea of converting GRAIL into a data company because of the potentially high value of its data and the low value (and cash incinerating ability) of it diagnostics test (e.g. a year ago https://lnkd.in/gr4uc-rn). It's a great time to revisit this idea for two reasons: (1) The EU is forcing Illumina to divest GRAIL in the exact same form as it was acquired in - a MCED test. But it looks like once GRAIL is spun out there are no further EU restrictions - freedom! GRAIL can morph into whatever it wants to be, including halting the huge burn of Galleri sales and marketing and flipping to data sales model. (2) We can now see that Tempus AI has $170M in fast-growing revenue from DATA which it defines as "the licensing of de-identified datasets to third parties and by providing clinical trial support, such as matching patients to clinical trials enrolled in its clinical trial network." It's a real business! And better yet it has a 67% gross margin vs 48% for testing 😍 Some differences. Tempus has spent a total of about $1B building its database, GRAIL has spent about $4B. But that's an amazing advantage (yep!) for GRAIL stock owners (ie soon all Illumina stock owners) in the context of a data strategy: GRAIL investors and ILMN paid $4B for a database that you will now get for about $0 with $1B thrown in to build a business around it 👏👏👏 Two futures for GRAIL: (1) Current path: company keeps its current strategy of struggling to sell a low performing MCED test that patients (and payers) aren't interested in paying for. GRAIL burns down its $1B Illumina gift and goes the way of Invitae 🤯 (2) Data path: GRAIL recognizes that it has invested $4B in building an incredible (genomics + methylation + EMR) database + analytics tools and focuses on monetizing that data and adding to it with other omics (e.g. proteomics). Data customers include both traditional biopharma and the new generation of LLM-based drug developers hungry for training data. Ditching Galleri (by my calculations, not an analyst 🤣) would give GRAIL a burn of about $240M vs current $700M - so a much, much longer runway to build this business 👏 My bias is clear I think 🤣🤣🤣

I have a ton or respect for Alex's commentary but I'm just not there yet on the data business models. With an echo of Michael Porter in my ears, what is the sustainable economic moat for healthcare or patient data models? Seems to be relatively little other than time to prevent other groups from replicating much of what Tempus has generated. I get there are willing buyers of Tempus data today but without some type of moat the value of that data should be priced like a commodity (not unlike equity research) and decline as supply increases. Diagnostics are a horribly difficult business model due largely to byzantine reimbursement but it is a business with real barriers to entry (regulatory/clinical utility datasets/payor contracts). With clear reimbursement guardrails (huge challenge for Grail), diagnostic businesses can work. Market cap for EXAS is $10B+ and creeping up on ILMN. It's not that Cologuard is an amazing product. It's the moat.

Join our workshop for free! 9. The Near-Future of Regulatory Submissions: Embracing Reproducibility, Open Source, and Interactivity With Leena Khatri!!

From data chaos to drug design: a new platform speeds discovery AI-driven drug-discovery success relies on data quality — what happens when a century of scientific curation is added to the mix?

Join me VIRTUALLY at R/Pharma (FREE) on Friday, November 1st at 1PM EST for my workshop "Visualizing Clinical Trial Data: Foundational Principles for Data Visualization, Best Practices, and Programming Techniques in R by Example." 🎉 Unlock the power of visualization in clinical trial data analysis with this engaging workshop. Begin with an introduction to key conceptual practices from visualization experts like Stephen Few and Kirk Paul Lafler, then apply these principles to create compelling survival, failure, and swimmer plots, as well as other essential graphs using R by example. Drawing from the acclaimed “SAS Graphics for Clinical Trials by Example” by Harris and Watson, this session will show you how to transform complex data into clear, impactful visuals. Gain practical skills in utilizing R packages like ggplot2 and survminer to produce high-quality graphics that meet industry standards. #RProgramming #RStats #OpenSource #Pharmaverse #ClinicalTrial #ClinicalTrialData #DataViz #RPharma Sign up for the virtual workshop here: https://lnkd.in/gTVkHNhn

Day 6 of 100: 🚀 Just started a project where I am to integrate API for real-time predictions with machine learning models. It’s amazing to see how a seamless API connection can unlock new insights! Kedrus Academy Kedrion Biopharma #MachineLearning #DataScience #AIintegration

🚀 Exciting Announcement: New Publication in #AIToxicology! 🧬 I am happy to share that our latest #AIToxicology paper with Donghyeon Kim and Jaeseong Jeong has just been published! This work represents a significant step forward in the field of AI-driven toxicology, addressing a critical challenge in toxicity prediction: balancing model performance with interpretability. In this study, we explored the #ToxCastTox21 bioassay data set, analyzing 1092 assays using five different molecular fingerprints and six machine learning algorithms. Our goal was to find the optimal combination that not only delivers strong predictive performance but also provides explainable insights into toxicity-related chemical structures. This research underscores the importance of considering simple, interpretable models in #AIToxicology, especially when working with chemical feature-based data. By doing so, we can ensure that our models are not only powerful but also transparent and accessible for regulatory validation. https://lnkd.in/g6jrFbSx #AIToxicology #MachineLearning #Toxicology #ExplainableAI #ToxCastTox21 #ChemicalSafety #ResearchInnovation

OpenPhenom-S/16 from #Recursion is now available on #Googlecloud. Anyone can now explore and experiment with the same technology Recursion uses for phenotypic drug discovery. Read more about how Recursion is making this #foundationmodel publicly available here: https://lnkd.in/dyK_rS4q