Skip to content
Stay curious ↗

Follow your curiosity.

Try “brain”, “genomics”, or another subject you follow.

Today/Drug Development
Drug Development · The big question

The Missing Ingredient in AI Drug Discovery: Better Data

Why AI Alone Isn’t Enough for Oligonucleotide Discovery

GEN2 min readAI-assisted summary
Editorial illustration for Why AI Alone Isn’t Enough for Oligonucleotide Discovery
Editorial illustration

Summary

Artificial intelligence is often hailed as a magic wand for modern medicine. Researchers are especially hopeful about using AI to speed up the discovery of oligonucleotide therapies. These are short strands of lab-made genetic material designed to treat diseases at the genetic level. Accelerating this discovery process is vital for treating what doctors call "n=1" rare diseases, which are unique genetic conditions that might affect only a single patient in the world. For these individuals, finding a treatment quickly is incredibly urgent.

However, current AI tools built to predict how well these genetic drugs will perform are struggling with accuracy. Performance scores for models like ASOptimizer and OligoAI generally hover between a modest 0.4 and 0.7 correlation. The limitation is not the computer code, but the information feeding it. Right now, scientists mostly train these models using publicly available information. This public data is inherently messy. It comes from inconsistent laboratory conditions, frequently leaves out negative data from failed experiments, and lacks crucial information about toxicity and unintended side effects.

To overcome these early-stage hurdles, scientists are changing how they gather information. By integrating a technology called high-throughput RNA-sequencing into their screening process, researchers can systematically detect off-target effects and vastly improve the quality of their datasets. Instead of scraping public databases for flawed information, researchers are learning that they must generate their own massive, consistent datasets in the lab.

Why it matters.

For the biotechnology industry, this highlights a crucial reality: algorithms are useless without high-quality fuel. To actually realize the predictive power of artificial intelligence, pharmaceutical organizations must invest early in proprietary, large-scale screening operations and well-organized data infrastructures. Relying solely on software and public data is no longer a viable strategy for building the next generation of medicines.

This evolution in drug development is especially critical for patients with highly specific rare diseases. High-quality data ensures that AI models can accurately predict and avoid dangerous off-target effects before a drug ever enters a human body. While building these massive proprietary datasets requires significant time and money upfront, it is a necessary foundation. It is the only way to unlock the true potential of artificial intelligence and deliver safe, personalized medicines to the patients who need them most.

From the source

The story in numbers.

Correlation statistics between predicted and experimental values for current AI oligo models
0.4 - 0.7
Target patient scale for personalized rare disease oligonucleotide therapeutics
n=1
Keep the context

Follow the evidence.

This AI-assisted summary is a starting point. Read the original reporting for its methods, scope, and limitations.

Your next discovery.

Keep exploring ↗
Drug Development

A $122M boost for a fast, inside-the-body approach to cell therapy

Create Medicines recently secured $122 million to advance a new class of treatments for cancer and autoimmune diseases. They specialize in CAR-T cell therapy, a treatment that engineers the immune system to fight disease. But instead of the traditional method of removing cells from a patient to alter them in a lab, Create Medicines is developing therapies that work entirely inside the body, a process known as "in vivo" treatment. To achieve this, the therapy uses tiny fat bubbles called lipid nanoparticles to deliver messenger RNA, or mRNA, into the patient. Think of mRNA as a set of biological instructions. Once delivered, these instructions directly teach the patient's existing immune cells, including T cells, natural killer cells, and myeloid cells, to reprogram themselves into disease fighters. The startup has already dosed more than 50 patients, giving it the largest clinical dataset in the in vivo field. They recently began an early-stage clinical trial for an experimental therapy designed to attack solid tumors that carry a specific protein called HER2. Alongside this cancer research, the company is testing other treatments that target proteins called CD19 and BCMA to help patients with autoimmune conditions.

BioPharma Dive1 min read
Drug Development

Supreme Court Protects Access to Common Abortion Pill

The United States Supreme Court has decided to preserve access to mifepristone, a medication used in the most common method of abortion. By rejecting recent restrictions proposed by lower courts, the Supreme Court has ensured that patients can still obtain this widely used drug while an ongoing lawsuit continues to make its way through the legal system. Under this current legal order, patients do not need an in-person visit with a doctor to get the medication. Instead, they are permitted to continue receiving mifepristone directly from pharmacies or through the mail. This ruling maintains the existing distribution methods, meaning that access is expected to remain uninterrupted at least into the following year as further legal proceedings unfold.

STAT News1 min read
Drug Development

Living Implants: Engineered Bacteria Trapped in Gel Fight Infections

What if a medical implant could sense an infection and fight it off automatically? Researchers have created exactly that with a new technology called an "implantable living material." To build it, scientists used a specialized, water-based gel known as a polyvinyl alcohol hydrogel. This material acts as a tiny, flexible fortress. It is uniquely stiff and tough, designed to hold genetically engineered bacteria safely inside for up to six months without them escaping into the surrounding tissue. Inside this protective gel, the researchers trapped a modified version of E. coli bacteria to act as molecular guards. These engineered microbes are programmed to detect chemical signals given off by Pseudomonas aeruginosa, a dangerous pathogen that frequently causes infections around artificial joints. When the trapped bacteria sense the invader, they trigger a targeted self-destruct mechanism. This intentional cellular explosion releases a powerful antibacterial protein that easily passes through the microscopic pores of the gel to kill the infection outside. The research team tested this system in mice with artificial joint infections. Over a three-day treatment period, the living implant significantly reduced the amount of infectious bacteria compared to non-therapeutic controls. By keeping the helpful microbes physically contained while allowing their medicine to flow out, the researchers have found a clever way to turn living cells into localized drug factories.

GEN2 min read

A little science goes a long way.

Clear, thoughtful biotech briefings. In your inbox three times a week.

Stay curious. Subscribe free ↗
The Missing Ingredient in AI Drug Discovery: Better Data — Daily Strand