Friday, October 16, 2026

Google.org awards $2.3 million USD to University of Toronto researchers to advance AI-powered peptide drug discovery

Research team, led by Joseph Brown, will generate experimental data and develop AI model to create open access peptide research tool

University of Toronto - Leslie Dan Faculty of Pharmacy

New AI-powered OpenPep platform aims to accelerate peptide drug discovery — image:
From left: Chris Ing, co-founder ProteinQure; Joseph Brown, Assistant Professor, Leslie Dan Faculty of Pharmacy, University of Toronto; Philip Kim, Professor, Department of Molecular Genetics, University of Toronto; Keith Pardee, Associate Professor, Leslie Dan Faculty of Pharmacy, University of Toronto
view more
Credit: Steve Southon

A group of scientists led by Joseph Brown, assistant professor at the Leslie Dan Faculty of Pharmacy, has received $2.3 million USD in funding to develop a database and AI training model to identify peptides with therapeutic potential. The project is funded through the Google.org Impact Challenge: AI for Science initiative.

The newly funded initiative, OpenPep, will lead to an open access database that will allow researchers to use artificial intelligence (AI) to predict which peptides will bind to target proteins.

“Peptides are having a moment: Many adults have tried a peptide drug, and 8% of newly approved drugs are peptides. They are no longer a niche class of drugs,” says Brown, who is also a Scientific Advisor for the University of Toronto’s Acceleration Consortium. “The real challenge is that we don’t have a lot of existing data for peptide discovery, so developing peptide medicines is still remarkably slow. That challenge inspired us to gather large amounts of data so we can build chemically-informed models and speed up the pace of discovery.”

Peptides are short chains of amino acids, smaller than a protein but larger than a conventional small molecule drug. When engineered, they can be stable, precise and selective in modulating biological targets and they hold a lot of promise as drugs.

AI has become an important part of biological research and drug discovery, as researchers can leverage AI models where data exists to make predictions and speed up the research process by prioritizing which wet-lab experiments to complete. However, data is the foundation from which AI is built. For example, AlphaFold, one of the most prominent AI models for protein structure prediction, was built from open-source data developed over decades of research.

But no similar database and AI model exist for peptides. Peptide discovery almost exclusively progressed by experiments, with the precious data scattered across various labs, projects, and companies.

To tackle this gap, Brown and his co-investigators – Associate Professor Keith Pardee from the Leslie Dan Faculty of Pharmacy, Professor Phil Kim from U of T’s Donnelly Centre for Cellular and Biomolecular Research, and Chris Ing, co-founder and chief scientific officer of ProteinQure – decided to team up to build a database and matching AI model for peptide binding that researchers around the world can use and contribute to.

“AI has the potential to transform how we understand biology and accelerate the discovery of new therapeutics,” said Leslie Yeh, Director of Scientific Progress, Google.org. “By pairing our targeted, early-stage funding with advanced AI, we can give researchers the runway to take bold risks on foundational science that traditional sources of capital may overlook. OpenPep aims to do just that—breaking down fragmented data barriers so scientists around the world can openly share knowledge and build on one another’s discoveries. We’re proud to support the University of Toronto in building this resource and advancing a more open, collaborative approach to AI-powered scientific discovery.”

Research team will crowd-source suggestions of disease-relevant proteins to target

OpenPep leverages the expertise of the co-investigators across peptide discovery, design, and AI/machine learning to generate large amounts of data and build models to predict peptide-protein binding.

Pardee’s group will leverage their expertise in accessible cell-free systems from synthetic biology to produce up to 500 disease-relevant proteins, a snapshot of the human proteome. Brown’s team will use specialized mass spectrometry to experimentally test large libraries of peptides against the target proteins and measure which peptides bind and their binding strength.

The team will then use the experimental data to build and train AI models that can predict how well peptides bind to proteins that have not been tested experimentally. As the researchers test and validate those predictions, the data can be added to the model to further refine and improve predictions.

“AI can turn thousands of individual peptide experiments into generalizable rules about how peptides interact with proteins,” says Ing. “With OpenPep, the goal is to use those insights to predict and design peptides for targets we haven’t experimentally explored yet, allowing the value of each experiment to extend well beyond the specific peptide and target that were measured.”

As the team starts working on generating experimental data, they will be crowd-sourcing suggestions of proteins for peptide discovery from collaborators at hospitals, pharmaceutical companies, and research institutes with particular interest in pediatric and rare diseases, cancer, neurodegeneration and cardiovascular disease. Any researcher interested in nominating a target for discovery can submit suggestions via a webform found on the team’s website openpep.com.

Once established, OpenPep will be an open database that the team and other researchers can continue to contribute toward further model improvements. “Some of the biggest advances in biology have been built on open scientific resources. No single laboratory can explore the full diversity of peptide chemistry and disease biology, so making the resource open also gives it the potential to grow through contributions from the broader scientific community,” says Ing.

“OpenPep will give peptide drug-discovery teams around the world a head start. Researchers will be able to build on a large body of experimental data to find and optimize promising peptide molecules more efficiently. Ultimately, we hope this makes peptide discovery more predictive and shortens the path from a disease target to a potential therapeutic.”

Brown says the significant investment from the Google.org Impact Challenge Award is key to ensuring that the researchers are able to deliver on their ambitious goal.

“This funding gives us the momentum to deliver on committing to dedicated data generation efforts required to advance the field in this way,” says Brown. “The scale of data that we’re going to put together doesn’t exist anywhere else, and this support is going to help us lead in the space and unlock a completely different AI-powered paradigm for peptide discovery.”

“AI won’t replace experiments, but it will help us choose better ones by identifying the peptides most worth testing,” says Brown. The team further states that “by turning experiments into open data, and open data into AI, we hope to give more scientists a better place to start --- and accelerate the discovery of new peptides for the benefit of everyone.”

CA Skip navigation Search Create Avatar image OpenPep: A Breakthrough for AI-Driven Peptide Drug Discovery [VIDEO]

Most patients won't endorse AI medical responses without a clinician's review, Columbia Nursing study in JAMA Network Open finds

Columbia University Irving Medical Center

NEW YORK, NY (October 14, 2026) — In a Columbia University School of Nursing study surveying 1,185 U.S.-based adults about AI use in secure medical portals, 63% of patients said they would not endorse AI-generated medical responses without a clinician's review. This suggests that while patients see value in using AI for certain health care tasks, they prefer a human review when medical advice is involved. The study, “Public Perceptions of Timeliness for AI Use and Human Oversight,” was published in the Journal of the American Medical Association (JAMA) Network Open on October 14, 2026.

Researchers found that patients' preferences differed by task. Most respondents preferred fast AI-generated responses for logistical requests such as appointment scheduling and prescription refills (57%). However, two-thirds of participants said they would rather wait up to 48 hours for a doctor or nurse to review and edit an AI response to medical questions.

The researchers also examined attitudes around mental health support using AI. Participants said they mostly likely would use a licensed human therapist, even with a three-month wait (64%). Acceptance declined for a therapist-monitored AI chatbot available within one month (50%) and an AI-only chatbot available immediately (44%).

“This study showed what we had already assumed but can now back up with data. When people have a medical question, they want to hear from their doctor, not AI, even if AI may give an immediate answer,” said Natalie Benda, PhD, lead author and assistant professor at Columbia Nursing.

“People’s opinions differed across characteristics like race, health literacy, gender, and experiences of discrimination in medical settings. Notably, individuals who reported having been discriminated against in a medical setting were more open to AI responses to medical questions.”

The study highlights the importance of allowing patients to make decisions about how and when AI is used in their medical care. The authors suggest that patients may be more receptive to AI in lower-stakes situations but still expect clinician involvement for higher-stakes medical decisions and guidance.

Other study authors include Uday Suresh, PhD, Megan E. Salwei, PhD, and Laurie Lovett Novak, PhD, Vanderbilt University Medical Center; Afra Shamnath, MPH, and Meghan Reading Turchioe, PhD, Columbia University School of Nursing; Isabella Buitron, MPH, Columbia University Mailman School of Public Health; and Xuhai (Orson) Xu, PhD, Columbia University Department of Biomedical Informatics.

About Columbia University School of Nursing

Columbia University School of Nursing is advancing nursing education, research, and practice to advance health for all. As one of the top nursing schools in the country, we offer direct-entry master’s degrees, advanced nursing, and doctoral programs with the goal of shaping and setting standards for nursing everywhere. And, as a top recipient of NIH research funding, we address health disparities for under-resourced populations and advance equitable health policy and delivery.

Through our expansive network of clinical collaborations in New York City and around the world—including our dedicated faculty practice, the ColumbiaDoctors Nurse Practitioner Group—we cultivate a culture of innovation and diversity and champion a community-centered approach to care. Across the Columbia Nursing community, we encourage active listening, big thinking, and bold action, so that, together, we’re moving health forward.

Columbia University School of Nursing is part of Columbia University Irving Medical Center, which also includes the Columbia University Vagelos College of Physicians and Surgeons, the Mailman School of Public Health, and the College of Dental Medicine.

Journal

JAMA Network Open

DOI

10.1001/jamanetworkopen.2026.39048

Article Title

Public Perceptions of Timeliness for AI Use and Human Oversight

Article Publication Date

14-Oct-2026

UBC researchers uncover new cyber threat to water and energy networks

Hackers could identify and block vital communications to essential infrastructure without ever breaking its encryption, researchers warn

University of British Columbia

University of British Columbia researchers have uncovered a vulnerability that could allow hackers to target and disrupt critical infrastructure such as water treatment plants and energy grids without needing to break into a facility’s own systems or crack its encryption.

The technique targets facilities that communicate with cloud-based monitoring and control systems over encrypted connections. An attacker would first need to compromise a router carrying traffic between the facility and the cloud.

The findings suggest that encryption alone may not be enough to protect vital public facilities. By analyzing everyday traffic patterns like the timing and size of digital messages, an attacker could identify when critical commands or data are likely to be transmitted, even when their actual contents remain hidden.

Industrial control systems rely on computers to run physical machinery such as pumps, valves and generators. Unlike attacks that compromise equipment inside a plant or overwhelm a network with traffic, the new technique uses patterns in encrypted communications to predict critical moments and selectively block a small fraction of messages. This allows it to disrupt operations without entering the plant’s own systems and evade the attack detectors tested by the researchers during the attack.

“What makes this attack concerning is that an attacker doesn’t need to read the messages being exchanged,” said Dr. Aastha Mehta, a UBC computer science professor who leads the Principled Systems Security Group and is a co-investigator on the project. “Even when the contents are encrypted, patterns in the communication can reveal enough information to identify and target critical communications.”

How a ‘sniper’ could target a water-treatment plant

The researchers call their attack framework “ICS-Sniper” because, like a sniper, it can selectively target critical communications with a small footprint.

In laboratory tests, the team successfully demonstrated the attack on a cloud-adapted version of a water-treatment research testbed at Singapore University of Technology and Design. They used two models of the testbed, one using software-based controllers and the other using standard industrial control hardware; both models simulated the physical processes.

In one test, the attack delayed the start of clean-water pumping by about 15 minutes. Another left a simulated valve open for more than nine minutes longer than intended — a failure that could cause treated-water loss in a real plant.

“An attacker doesn’t necessarily need to crack your encryption to cause disruption,” said Dr. Karthik Pattabiraman, a UBC professor of electrical and computer engineering and co-investigator on the project who leads the Dependable Systems Lab. “By simply watching traffic patterns, they can identify when critical messages are likely to be sent and selectively block them. That means blocking a small amount of traffic could trigger a much larger operational disruption.”

The researchers chose water treatment because public schematics were available to replicate the facility’s control systems. They also tested their traffic-analysis method across several industrial communication protocols, including a power-grid dataset, although the operational disruptions were demonstrated in the water-treatment models.

“As industrial systems increasingly connect to cloud-hosted services, their exposure to network-based attacks like ICS-Sniper could grow,” said first author Dr. Gargi Mitra, who conducted the research as a postdoctoral researcher at UBC. “Our goal is to uncover new vulnerabilities before attackers do, so the security community can address them while these services are still in the early stages of adoption.”

Making critical systems harder to disrupt

The team is already working on ways to safeguard critical infrastructure against this type of attack.

One approach involves making network communications less predictable — for example, by padding message sizes or adjusting their timing to make it harder for attackers to distinguish critical commands from routine system chatter. Such changes would need to preserve the timing requirements of industrial equipment. Another is to make plant controls and coordination more resilient to network disruptions, including by providing alternative routes for critical communications.

“Our goal is to build defence in depth and make these systems more resilient to attacks like this,” said Dr. Mehta.

The team has also released its research code, instructions for reproducing the laboratory testbeds, and network traffic data from those testbeds so other security researchers can reproduce the findings and develop and test defences.

As society relies more heavily on interconnected technology to run essential services, the researchers say security strategies need to protect not only the privacy of data, but also the safe operation of physical machinery during an attack.

The paper, “ICS-Sniper: A Targeted Blackhole Attack on Encrypted ICS Traffic,” has been accepted for presentation at the 2027 Network and Distributed System Security Symposium, a leading computer security research conference, in Seoul, South Korea.

Chanyuan Liu, Pritam Dash, Elaine Yao and Alain Zhiyanov also contributed to the paper.

The project is supported by the National Cybersecurity Consortium.

Interview languages: English (Mehta, Mitra, Pattabiraman)

Method of Research

Computational simulation/modeling

The science behind the world’s fastest autonomous racing drone

Delft University of Technology

What does it take for an autonomous robot to be faster around a course than human drone-pilots? At these dizzying speeds, humans need lightning-fast reflexes. Researchers at Delft University of Technology (The Netherlands) have developed an AI system that enables a racing drone to navigate demanding courses at extreme speed using only a single camera. It outperformed even former world champion human drone-pilots in a race. Now published in Science Robotics, the scientific work behind this accomplishment shows how compact, efficient neural-network-based control can achieve high-performance autonomous flight under real-world conditions. The underlying principles could help advance a much broader class of autonomous robots, from vehicles to mobile and humanoid systems.

Less is more

The TU Delft team set out to investigate how far artificial intelligence can push a drone with very limited sensors, onboard computing power and energy. This is a challenging problem: at high speed, even a small delay in sensing or control can mean missing a gate, colliding with an obstacle or losing control and crashing.

The result is MonoRace, an autonomous flight system developed at the MAVLab of TU Delft's Faculty of Aerospace Engineering. Unlike previous leading autonomous racing systems, MonoRace uses just one forward-facing camera and an inertial measurement unit (IMU) for perception, navigation and control.

Seeing the track with one camera

At high speed, the drone has very little time to interpret what it sees. A neural network identifies the racing gates in the camera images, while the onboard IMU and a model of the drone estimate its position and motion. The system does all this without GPS or external motion tracking.

“Flying at very high speed with only one camera is challenging because the images are blurry, deformed and the AI system has very little time to precisely determine where the drone is,” says Stavrow Bahnam, main author of the study. “We developed a robust way of combining what the camera sees with the drone’s own motion information, allowing it to navigate the track at high speed without GPS.” The system also had to cope with imperfections that occur in real flight, including camera interference and sensor saturation.

Neural network takes direct control

The researchers also redesigned how the drone is controlled. Rather than using a conventional control system with several layers of intermediate controllers, MonoRace uses a small neural network that directly generates the commands for the drone’s motors.

The network runs onboard a tiny computing chip, the “flight controller”, giving control commands 500 times per second. It was trained using reinforcement learning, allowing it to learn how to control the drone while taking its physical limitations into account.

“Instead of first planning a desired flight path and conventional commands, our neural network learns to directly control the motors,” says Robin Ferede, author of the study. “This allows the controller to react extremely quickly and exploit the capabilities of the drone while remaining small enough to run onboard.”

Tested under extreme conditions:

To validate the system under realistic conditions, the researchers put it to the test in a demanding international competition environment. The 2025 Abu Dhabi Autonomous Drone Racing Competition provided a particularly rigorous challenge. The drone and racing environment were designed and operated by the competition organisers rather than by the Delft researchers. The system therefore had to operate successfully on both hardware and a track that the research team did not control. Despite these constraints, the TU Delft drone reached speeds of up to 100 km/h, defeating all other autonomous teams as well as three former world champion human pilots.

“Instead of giving the robot increasingly powerful sensors and computers, we wanted to find out how much intelligence we could extract from very limited resources,” says Christophe De Wagter. “Drone racing provided an extreme environment in which to test these ideas outside the laboratory. It pushed us to develop AI that was not only fast but also robust, and able to adapt when external reference measurements were unavailable.”

Bringing AI to the real world

For Guido de Croon, professor at Delft University of Technology, the value of drone racing goes beyond making drones fly faster. “Autonomous racing is a way of pushing robotic AI to its limits. In the current study, we made an AI efficient enough to control a drone travelling at 100 kilometres per hour using a single camera and a small onboard processor. The developed AI could be applied to many more, different real-world robotic systems. Here, we optimised for speed, but the same approach could be used to prioritise efficiency or safety. For example performing smooth, energy-optimal control on systems ranging from small vacuum cleaning robots to huge ship-to-shore cranes.”

The researchers believe that the same principles could help a wide range of autonomous robots make fast, reliable decisions with limited onboard resources, from vehicles and mobile robots to future aerial and humanoid systems.

Closeup of the TU Delft autonomous racing drone

Closeup shot of the TU Delft autonomous racing drone during the race in Abu Dhabi in 2025

Credit

Micro Air Vehicles Lab | Delft University of Technology

An autonomous racing drone flies through a gate during trials

The autonomous racing drone MonoRace flies through a gate during trials

Credit

Micro Air Vehicles Lab | Delft University of Technology

Caption

Timelapse image of the autonomous MonoRace racing drone flying through final racing gate at Abu Dhabi in 2025

Credit

Micro Air Vehicles Lab | Delft University of Technology