Biobanks: The Overlooked Key to Advancing AI in Pharma Research
ALSO: Plasma proteome for measuring organs aging; controlling ChatGPT with thoughts; Alphabet of AI in Healthcare; using bacterial communities to do useful stuff; and more
Hi! I am Andrii Buvailo, and this is my weekly newsletter, ‘Where Tech Meets Bio,’ where I talk about technologies, breakthroughs, and great companies moving the biopharma industry forward.
If you've received it, then you either subscribed or someone forwarded it to you. If the latter is the case, subscribe by pressing this button:
This week’s newsletter is brought to you by BiopharmaTrend, a well-established life sciences media portal and a technology scouting agency.
Disclaimer: This is my company (surprise!)
BiopharmaTrend has just launched its landmark report, “The State of AI in the Biopharma Industry,“ which is a 150-slide exploration of the current state of predictive and generative models in various aspects of drug discovery and development, synthetic biology, and biomanufacturing.
The report package also includes yearly access to our interactive database of 450+ AI companies in the domain of life sciences. The database is regularly updated in a semi-automatic way. It means that some data, like clinical trials, patents, and papers, is updated automatically by our engine, and other types of data, like funding or R&D collaborations, is added manually by someone from our team.
Learn more about “The State of AI in the Biopharma Industry“ and how it can assist you in strategizing, market research, VC deal sourcing, consulting, decision-making, preparing for deepteck keynotes, and more.
Last but not least, the report package provides some access to our team of analysts (myself included, sometimes) throughout the license period, which means we can do some free follow-up research upon agreement and answer some follow-up questions on the subject matter.
Now, let’s get to this week’s topics!
In a recent commentary for Drug Discovery World, Dr. Niven Narain, President and CEO of BPGbio, highlighted the crucial role of high-quality biobanks in enhancing AI-driven drug discovery and development.
It is not a surprise that the pharma/biotech industry is in dire need of high-quality data to train more useful deep learning models. Garbage-in, garbage-out is a well-known thing in the realm of artificial intelligence research as applied to drug discovery and biotech. But it seems like the topic of biobanks has been somehow overlooked in this context.
I’ve been tracking the progress of Berg Health, a Boston-based company acquired in February 2023 by BPGbio, since probably 2017. This clinical stage biotech was co-founded in 2006 by real estate billionaire Carl Berg and has been directed ever since by Niven Narain, who eventually became CEO of the BPGbio parent company. By 2014, Berg Health had grown to a 200-person team and already had a Phase Ib asset, while also collaborating in systems biology with academic and government groups. They marketed AI as a central paradigm already in 2014, when almost no one in the pharma industry seriously considered AI as a strategic priority.
Another thing that apparently differentiated the early Berg Health competitive position was the serious focus on biobanks and working with patient samples, getting multiomics studies done at scale, and modeling healthy vs. diseased states of cells and tissues.
Coming back to the present commentary, Narain addressed the varying utility of biobanks. While consumer genomics companies like 23andMe offer vast sample collections, they often lack detailed clinical and phenotypic data, crucial for understanding the nuances of diseases. In contrast, companies like BPGbio utilize multiomics data from private biobanks, enriched with comprehensive demographic, clinical, and experimental data. This approach is seen as essential for the success of AI-driven drug discovery.
Furthermore, the diversity and quality of biobanked samples are vital. BPGbio's biobank, which boasts over 100,000 patient samples, benefits from its partnerships with the U.S. Department of Defense and various medical and academic institutions. This diversity in samples aids in creating more effective AI models for drug discovery.
BPGbio has leveraged these high-quality biobanked samples and its proprietary AI model, run on the Frontier supercomputer, to identify diagnostic markers and develop multiple drug compounds. These are currently in Phase II studies for diseases like glioblastoma multiforme and pancreatic cancer, alongside programs for other conditions including epidermolysis bullosa and squamous cell carcinoma.
Narain believes that addressing the translational challenges in drug discovery will require scaling up biobanking efforts. This involves collecting more diverse samples with richer clinical annotation to refine AI models. He envisions that such advancements will pave the way for more personalized and optimized medicines, tailored to specific population subsets.
This future state, according to Narain, necessitates a concerted effort across the industry to expand biobank collections and utilize them effectively in AI-driven drug development.
Pick of the Week
There is an impressive study, Organ Aging Signatures in the Plasma Proteome Track Health and Disease published in Nature.
This study utilized blood plasma proteins to assess organ-specific aging in humans, analyzing aging in 11 major organs across 5,676 adults using machine learning. The findings revealed that 20% of individuals exhibit accelerated aging in one organ, with multi-organ aging in 1.7% of the population, correlating to a 20–50% increased mortality risk.
Particularly, accelerated heart aging was linked to a 250% increased risk of heart failure, and rapid aging of the brain and vascular systems was a strong predictor of Alzheimer's disease, comparable to plasma pTau-181, a leading biomarker for Alzheimer's.
All in all, being able to measure and quantify aging is crucial for being able to actually measure possible longevity interventions or lifestyle choices.
First True Telepathy With a Machine
MindPortal, a tech company merging artificial intelligence (AI) and neurotechnology, has recently unveiled a quite futuristic, non-invasive optical brain-computer interface (BCI). This technology enables direct communication between human thoughts and AI, specifically with ChatGPT, the AI-powered large language model developed by OpenAI.
Ekram Alam, CEO and Co-Founder of MindPortal, emphasizes the significance of this development:
"For the first time, we have successfully enabled telepathic communication between humans and artificial intelligence."
Brain-computer interfaces, like the one developed by MindPortal, work by analyzing brain signal activity in real-time. These signals are then translated into user commands, enabling control over external devices. This technology offers some hope, particularly for individuals with disabilities, as it allows for control of robotic limbs, speech synthesizers, and computing devices purely through thought.
Unlike invasive counterparts such as Elon Musk’s Neuralink, which requires a surgical procedure for implantation, MindPortal's BCI utilizes non-invasive optical sensors. These sensors leverage AI machine learning to decode full sentences in real-time rather than just individual words, marking a potentially significant advancement in the field.
Digging Gems
In my ‘Digging Gems‘ category, I select 1-2 things that are not just interesting tech+bio news, but something more than that. It might be a beautiful concept, a great design, an engaging story, or a particularly useful tool. Today is no exception.
I’ve stumbled upon this really creative idea by the team at OWKIN, a Paris and New York-based AI company, who created an “Alphabet of AI in Healthcare.” It is educational material explaining key tech concepts, and it is also a great marketing tool for OWKIN, as each letter and the story behind it would also promote a related service by the company.
The next gem for today is this video explainer by S3. Not only do I love the way Jason Carman records videos, but the company in this video, Concerto Biosciences, is also notable for its technology, kChip.
kChip physically constructs millions of miniature microbial communities by combinatorially recombining a library of microbes and/or environmental factors. By observing each constructed community, scientists at Concerto can identify key ecological ingredients—microbes, prebiotics, and postbiotics—that drive the beneficial capabilities of natural communities.
The insights from such experiments enable the company to develop transformative microbial, prebiotic, or postbiotic products in partnership with other companies.
I suggest exploring more videos from S3.
Weekly Highlights
U.S. approves first gene-editing treatment, Casgevy, for sickle cell disease The biggest news of the week, for sure. The FDA granted approval of Casgevy to Vertex Pharmaceuticals Inc. and approval of Lyfgenia to Bluebird Bio Inc.
The weird thing is that the stocks of Bluebird Bio fell around 40% on the news of approval. The explanation I found was due to a black-box warning and because of a higher price point than a competitor's treatment. But I honestly don’t get it anyway. 40% crash is by no means what you expect when the FDA grants authorization for something.
Disclaimer: This is not financial advice, and I am not an expert in biotech stocks. Consult experts when making investment decisions at all times (consider consulting a psychotherapist when investing in biotech).
A New Drug That Could Extend Dogs’ Lives Inches Closer to Approval
The FDA has expressed initial support for approving a longevity-focused drug, a first in regulatory history.
New approach to stem cell transplantation may overcome immune rejection
Baoyang Hu and his team at the Chinese Academy of Sciences have engineered stem cells to create immune-evasive liver cells, mitigating the risks of immune rejection and tumor development.
ISM5411 is an oral, PHD 1/2-specific inhibitor that facilitates the treatment of IBD by inducing the expression of gut barrier protective genes.
ISM5411 is Insilico Medicine’s fifth AI-driven drug program to enter clinical trials, with the first dose of healthy subjects completed in Australia.
To further evaluate ISM5411 in wider populations, Insilico plans to conduct global, multi-center Phase Ib trials in patients with ulcerative colitis following the Phase Ia study.
Whole Human Brain Neuro-Mapping at Cellular Resolution on NVIDIA DGX
AI Training AI: GatorTronGPT at the Forefront of University of Florida’s Medical AI Innovations yet another example of using synthetic data to train medical AI models
The power of prompting ChatGPT a great article demonstrating how prompt-engineering is a really powerful thing that may compete even with fine-tuning foundation models.
Molecular science VR startup Nanome launches AI copilot MARA
Nanome, a San Diego-based startup known for its innovative approach in enabling scientists to visualize 3D molecular structures through virtual reality, has recently unveiled its new AI-driven assistant, MARA. Designed to function like ChatGPT, MARA is accessible via a web interface and acts as a supportive tool for chemists, particularly those engaged in biopharmaceutical research and development. This AI copilot is equipped to perform standard cheminformatics tasks and deliver detailed answers to scientific inquiries, thereby enhancing the research process, according to the company's announcement.
Discovery of a First-in-Class Small-Molecule Ligand for WDR91 Using DNA-Encoded Chemical Library Selection Followed by Machine Learning The power of DNA-encoded libraries!