Despite its futuristic sound, artificial intelligence (AI) is already present in many modern devices. For instance, it enables voice and facial recognition on our mobile devices.
In biotechnology, where it has proven crucial to many facets of drug discovery and development, AI is also starting to become more noticeable.
Drug target identification, drug screening, image screening, and predictive modeling are all examples of AI uses in the biotech industry. Additionally, clinical trial data is managed and the scientific literature is searched using AI.
AI can manage various clinical trial datasets, offer virtual screening, and evaluate enormous amounts of data by utilizing machine learning.
AI can get previously unattainable insights and feed them back into the drug development process in addition to lowering the expenses of clinical trials.
Many businesses are developing AI technology to benefit the biotech sector. As outdated techniques like manual picture scanning and classical statistical analysis reach their practical limits, their services are quickly becoming indispensable.
A new world of Abundance:
Convolutional neural networking (CNN), a popular form of machine learning, was applied to drug design and discovery for the first time by Atomwise. CNN is utilized in well-known, commonplace applications like the image tagging function on Facebook or the speech recognition technology in Alexa.
On issues including hit identification, potency optimization, selectivity optimization, and off-target toxicity testing, Atomwise has 550 active machine-learning projects.
There is essentially no limit to the amount of small-molecule compounds that may be visually screened using Atomwise’s algorithm, said Abraham Heifets, Ph.D., CEO of Atomwise.
Heifets claims, “We recently completed the largest screen in human history—12 billion molecules. Most of the 12 billion molecules don’t and never have existed in nature. However, Atomwise partners are able to synthesize any of them.
These suppliers can provide intriguing candidate compounds in four to six weeks from a collection of fundamental building components.
Heifets claims that when those hypothetical substances become more readily available, it may soon be possible to screen 100 billion molecules.
With a lot of compounds, the issues with drug screening change. “It sounds fairly fantastic if you have 100 billion molecules and a 99 percent accurate model.”
Heifets clarifies. But if you have a 1% false-positive rate, there will be a billion false positives for your correct response. In reality, you need computational tools that are 99.999 percent accurate or more to work fruitfully and successfully in this new world of abundance.
AI is causing a lot of enthusiasm, just like many other revolutionary breakthroughs in pharmaceutical research, like CRISPR gene editing, proteolysis that targets chimera-induced protein degradation, and RNA interference.
Heifets asserts that “artificial intelligence promises to [assist drug developers] pursue once intractable targets.” He makes the observation that extraordinary health outcomes frequently precede early interest in disruptive technologies like AI. He emphasizes that it holds the prospect of creating new opportunities.
Autonomous AI in real-world use:
The field of clinical diagnostics also has intriguing prospects for machine learning. For instance, Eyenuk is creating AI technology for use in the medical field. EyeArt®, the business’s first product to hit the market, uses AI to identify diseases from retinal pictures.
The sensitivity for detecting diabetic retinopathy was over 95% in a clinical trial that involved 942 patients and was conducted in 15 medical facilities across the United States.
Machine learning was used in the creation of EyeArt to train its algorithms on about 2 million photos. Imagine teaching your resident using that many visuals, says Kaushal Solanki, Ph.D., CEO of Eyenuk. That is simply not possible.
Everyone with diabetes should have a yearly test for diabetic retinopathy, according to experts. The United Kingdom is currently the only nation in the world to screen more than 80% of its diabetic population.
That equates to about 2.5 million patients whose retinal scans would require individualized expert assessment each year.
In a health technology review conducted by the National Health Service (NHS) of the United Kingdom, EyeArt was contrasted against a number of rival technologies.
EyeArt was judged to be significantly superior. The results of the evaluation, which were released in 2016, demonstrated that EyeArt has a sensitivity of 99.6 percent for detecting proliferative illness and a sensitivity of 93.8 percent for referable disease.
In order to use EyeArt for its screening programs, the NHS is now changing its workflow. whereas pilot tests are being conducted at six centers in the United Kingdom have been completed.
Channeling the data deluge:
With simple, homogeneous data, traditional methods of data analysis for drug development perform well. When the data is complex, however—for instance, when patient records detail several diagnoses, comorbidities, complex treatment regimens, and numerous meetings with clinics and clinicians—those algorithms fall short.
With the integration, analysis, and production of stratified patient groups via artificial intelligence. The planning and execution of clinical trials are being revolutionized by the capacity to manage complicated, multivariate data.
Leading this push for clinical data is Sensyne Health. Rabia T. Khan, Ph.D., the head of translational medicine at Sensyne, claims that the old methodology of drug discovery is unsustainable because it costs billions of dollars and still results in a high failure rate.
However, she adds that AI has the promise of lowering prices and failure rates. Sensyneis is collaborating with the NHS to collect patient information and make it possible to stratify patients for clinical trials.
Khan claims that artificial intelligence is necessary since the data is so noisy, sparse, and varied. Standard methods do not work to identify subpopulations in heart failure.
The ability to identify subpopulations of heart failure and demonstrate that there are more than just the two well-known subtypes of heart failure comes when you apply more sophisticated machine learning-based methodologies.
Actually, there are a number of subgroups. We are actively examining the effectiveness of medications in several of those various groupings. She believes that the industry will eventually switch from traditional randomized controlled trials to virtual trials.
Virtual trials will perform the heavy lifting thanks to artificial intelligence and provide a lot of the data that was previously only possible through pricey human trials. In actuality, this knowledge will be accessible for a potential medicine before it is ever tested on humans.
We will start with real-world data and correlate that to patient samples rather than “taking it from an abstract concept in a dish all the way through to clinical practice,” according to Khan and use that for medication development, and then feed the same information back into the clinical study.”
Precision Medicine Group is another business specializing in the administration of clinical trial data. QuartzBio, an AI platform that analyses biological and clinical data streams to extract information and insights to speed up drug development, was recently purchased by the company’s Precision for Medicine division.
There have been numerous attempts in clinical trials to direct data from various sources to clinical trial investigators so that they could rapidly and flexibly analyze the data holistically.
According to Cliff Culver, senior vice president of Precision Medicine Group, these approaches are rarely successful.
He asserts that “all of the information is separately generated and existing in fragmented formats.” For a pharmaceutical company, numerous individuals would spend weeks or months putting all of that together, especially when the focus is on bringing quantifiable reportable back to source data—like pictures or sequencing data—to enable ongoing quality control.
“As a result, analysis is noticeably put off, usually until after a trial is over. Furthermore, extensive data integration across trials within an organization is rarely possible due to a lack of bandwidth.
As a trial progresses, we do it to give the business regular updates on what is happening, and later, at the enterprise level, to make the most of the data.
Some of the methodologies made possible by the QuartzBio platform superficially resemble the unguided artificial intelligence studies employed by Google, Netflix, or other significant technological firms. Finding biological insights differs from picking a movie you would enjoy, though.
The data sets used in drug research are frequently smaller and less dynamic, and Culver says there is a pressing need to go beyond correlation in order to comprehend what is happening biologically.
Our unique selling point is our capacity to compile all of that information, giving you access to the broadest data set for analysis and subsequent data-driven decision-making or computational biology analysis, the AI, to derive meaning that can be put into practice.”
Precision medicine business Concerto HealthAI has a strong oncology focus. It reveals how patients react to therapies in actual settings using artificial intelligence and machine learning.
The activity of the company can direct pharmaceutical research, enlighten outcomes research and value-based studies, and hasten the creation of new drugs.
Jeff Elton, Ph.D., the CEO of Concerto HealthAI, used the example of a typical cancer patient receiving their initial diagnosis. That patient would initially be staged according to the location and size of the tumor if it had spread, and other characteristics of the illness.
Then, the treatment would be directed by this knowledge. Then, more details—including “on the fly” data regarding the patient’s development or adjustments to the disease’s state—might become accessible. But it won’t instantly become a part of anything of the patient’s electronic medical record.
For pharmaceutical businesses, this kind of real-time information is essential. Pharmaceutical companies, for instance, can better understand why a medication is or isn’t effective by having the correct information on staging at various times in the patient’s history.
According to Elton, “We build artificial intelligence models that read the record, compute a stage, and provide an accuracy score based on everything that is in the record—imaging reports, molecular reports, We even configured analyses to only consider data that satisfies a specific accuracy criteria.
The technology from Concerto HealthAI can also be used to anticipate outcomes, such as whether a patient will respond or whether the response will be sustained. These forecasts may be helpful in planning a clinical investigation.
They can also give clinical researchers a clear view of the standard of care, which is crucial because frequently doctors won’t enroll patients in trials if they are too burdensome in comparison to the standard of care. For the purpose of trial design, Concerto HealthAI’s algorithms can forecast patient burden.
According to Elton, all these tools enable researchers to perform in-the-moment analysis, which would have normally needed weeks of data preparation.