Artificial intelligence is being used to speed up trials
Machine Learning for Health Research: A Case Study of Structural Changes in Infants at MoBa and an International Trial Pathfinder
The National Institute of Environmental Health Sciences in Durham, North Carolina was where Hberg joined a group. She contributed to a team that analyzed 1,054 samples from MoBa and identified 10 genes altered in infants born to mothers who smoked while pregnant. The 2012 study gave important evidence that smoking can cause subtle changes to the genome that do not alter the sequence of the DNA. “We are only beginning to understand the gravity of epigenetic changes during development,” says Håberg.
Håberg is passionate about connecting specialists from her team with interdisciplinary groups from around the world so that they can explore large amounts of data that hold clues about fetal health. A project is comparing MoBa data with information from a birth cohort in the country. “It all comes down to finding exciting new ways for teams of specialists to work together,” she says. It is gratifying to see so much resources devoted to questions of early embryonic development. Amy Coombs is related to me.
Deep Learning is a machine-learning technique that can identify complex patterns. In testing, their system analysed images with variables such as glasses with greater accuracy and less human oversight than previous methods. She found that the machine-learning techniques were rapidly becoming more sophisticated and could handle more data, all without the traditional human reviewer.
In 2022, the authors of a paper described how artificial intelligence could be used to classify tumours in tissue slides. “With deep learning, we can detect patterns that the human eye cannot see,” she says.
The team showed how an artificial intelligence system could be trained to identify bladder cancer- related defects in samples stained with H&E. “We do not aim to replace the urologist, but deep-learning can offer additional analysis,” says Ghaffari Laleh.
AI can eliminate some of the guesswork and manual labour from optimizing eligibility criteria. Zou says that sometimes even teams working at the same company and studying the same disease can come up with different criteria for a trial. Several firms, including Genentech and AstraZeneca, are using a tool called Trial pathfinder. Sun’s lab in Illinois has developed a method to train a large language model so that a user can provide a trial description and then ask it to generate an appropriate criterion range for their body mass index.
Ghaffari Laleh hopes to apply her skills to help medical professionals in developing countries who cannot afford to run advanced diagnostics and who struggle to recruit and train skilled professionals. “AI is a much more affordable option,” she says. Hospitals that lack resources can use a deep-learning model to analyse patient data from a wide range of countries. She is trying to develop an artificial intelligence that can read text, echocardiograms andradiology image data to speed up the work of doctors and other specialists worldwide. Amy Coombs is related to this person.
Having the capacity to launch research projects quickly proved invaluable to Patalon and her team during the COVID-19 pandemic, when global treatment and vaccination protocols changed rapidly to keep up with the evolution of the disease. In 2021, as the highly contagious Delta wave was surging through Israel, Patalon co-led a team that scoured the health records of almost 125,000 Israelis, charting coronavirus incidence, symptoms and hospitalization rates over three months.
The team discovered that vaccinated people who had not previously tested positive for COVID-19 were 13 times more likely to be infected by the new variant, compared with previously infected individuals who were unvaccinated. The results showed that the SARS-CoV-2 virus confers natural immunity to those who have been bitten, providing important evidence that vaccination was not an immediate priority. It was a big achievement for us.
Extracting new insights from the vast amounts of public-health data that are being collected globally is key to advancing treatments and keeping one step ahead of infectious diseases, says Patalon. She has a role at Ksm where she is in charge of theTipa Biobank, which has one million samples of blood from 200,000 patients. In addition to one-off samples from patients, the biobank collects serial samples — successive samples from the same patient over a period of time. Serial samples are “very rare and highly valuable for research”, says Patalon, especially when it comes to analysing biological changes before and after a diagnosis.
The link between hunger and the brain: a case study of Patalon’s work with Oliver Sacks during the ‘Gaza War’
Patalon’s team has been heavily affected by the war in Gaza and he says that being a leader is crucial in times of crisis.
This is a time that requires patience, emotional support, and the building of good relationships. We have to come out of this situation stronger.” Sandy Ong is my cousin.
Sarah was introduced to the work of Oliver Sacks when she was a student at the University of Wisconsin-Madison.
She runs a lab at Singapore’s Agency for Science, Technology and Research (A*STAR) where she studies the connection between hunger and the brain to help patients with metabolic disorders. She first studied this connection as a postdoctoral fellow in an adjacent lab, where she was part of a team that discovered a mechanism that regulates feeding.
The environmental cues that can cause hunger can be dealt with by the other centres. In a series of follow-up experiments10, Luo and her colleagues observed that when mice were placed in the same feeding chamber where the neurons in the tuberal nucleus had been activated the previous week, they would immediately start eating, even if it was outside their normal feeding times. The results suggest that these neurons not only influence basic feeding behaviour, but also integrate memory and contextual cues into the eating process, says Luo.
“Your neurons might become activated, just because of the environment you’re in,” says Luo. Those signals could cause you to eat, even if you are not hungry.
“It would be very invasive to implant an electrode in the brain to activate or inhibit these pathways,” says Luo. It could be accomplished by using pathways that connect to the brain using a technique that involves implanting a pulse generator under the skin on the chest. It’s possible that there will be a route to develop therapies to target some of these diseases. — Sandy Ong
Find Your Patients Now: Matching Clinical Trials with Online Sources Using Artificial Intelligence, Natural Language Models and Saama
Some of the tools developed by Saama can be used to predict when trials will hit certain goals. All data from a patient can be combined with tools to assess outcomes. Moneymaker says that it isn’t possible to analyse a patient’s picture by hand anymore.
Helping researchers and patients find each other doesn’t just speed up clinical research. It also makes it more robust. Often trials unnecessarily exclude populations such as children, the elderly or people who are pregnant, but AI can find ways to include them. People with terminal cancer and those with rare diseases have an especially hard time finding trials to join. “These patients sometimes do more work than clinicians in diligently searching for trial opportunities,” Weng says. AI can help match them with relevant projects.
The company helps pharmaceutical firms submit clinical-trial reports for approval by the FDA. What the company calls its Intelligent Systematic Literature Review extracts data from comparison trials. Another tool searches social media for what people are saying about diseases and drugs in order to demonstrate unmet needs in communities, especially those that feel underserved. Researchers can add this information to reports.
Chatbots can answer patients’ questions, whether during a study or in normal clinical practice. In order to give questions and answers to other people, one study4 took questions and answers from the AskDocs forum. More than 70% of the time, health-care professionals preferred the answers to the doctors by ChatGPLt. Researchers created a tool called ChatDoctor by tweaking a large language model and giving it real-time access to online sources. The medical information that ChatDoctor was able to answer was more recent than the training data.
Once researchers have settled on eligibility criteria, they must find eligible patients. The lab of Chunhua Weng, a biomedical informatician at Columbia University in New York City (who has also worked on optimizing eligibility criteria), has developed Criteria2Query. Users can use natural language to type in the criteria for inclusion and exclusion, or enter the trial identification number into the web-based interface and the program finds matching candidates in patient databases.
A company called Intelligent Medical Objects in Rosemont, Illinois, has developed SEETrials, a method for prompting OpenAI’s large language model GPT-4 to extract safety and efficacy information from the abstracts of clinical trials. This enables trial designers to quickly see how other researchers have designed trials and what the outcomes have been. The lab of Michael Snyder, a geneticist at Stanford University in California, developed a tool last year called CliniDigest that simultaneously summarizes dozens of records from ClinicalTrials.gov, the main US registry for medical trials, adding references to the unified summary. They’ve used it to summarize how clinical researchers are using wearables such as smartwatches, sleep trackers and glucose monitors to gather patient data. Alexander is a computer-science student in the lab, and he says he has had a lot of conversations with people who have seen the potential but are not aware how to use them for the highest impact. “Best practice does not exist yet, as the field is moving so fast.”
The first step of the clinical-trials process is trial design. What should be given to the patient? How many patients are there? What data should be collected on them? The lab of Jimeng Sun developed an artificial intelligence that can predict whether a trial will succeed based on the drug molecule, target disease and patient eligibility criteria. When the training data is taken into account, SPOT also takes into account trials that took place more recently. Based on the predicted outcome, pharmaceutical companies might decide to alter a trial design, or try a different drug completely.
Computing power followed Moore’s law for a long time. The number of components on an integrated circuit doubled roughly every two years. The contrasting path of drug development was described by the term Eroom’s law. Over the previous 60 years, the number of drugs approved in the US has halved every nine years. It takes more than a billion dollars in funding and a decade of work to get a new medication to market. Half of the time is spent on trials which are larger and more complex. Only a small percentage of drugs enter phase I trials are eventually approved.
Researchers have developed an artificial intelligence (AI) machine learning system that can predict whether a clinical trial will fail based on the drug molecule, target disease and patient eligibility criteria. The machine learning system can extract safety and efficacy information from the abstract of clinical trials. It can also analyse patient data from a wide range of countries.