Artificial intelligence beyond the hype: achievements and expectations in the biotech space

Share this article

Artificial intelligence (AI) is a term that pops up everywhere lately. In this article, we provide some examples of how it opens a plethora of possibilities when it comes to improving healthcare. For example, AI is used in drug discovery, in diagnostics, in the analysis of clinical data and in telemedicine.

First of all, let’s get the terminology right. AI is an umbrella term for advanced computer techniques, developed since the 1950s, that include machine learning (ML) and deep learning (DL). ML creates algorithms and models to find patterns in data. It’s guided by predefined rules, which it uses as a basis to develop outcomes of a new dataset. Like ML, DL allows computers to solve complex problems through analyzing examples, yet deep learning takes this one step further. Deep learning builds the rules for pattern recognition in an automated way rather than through handcrafting, thereby making use of multi-layered neural networks, which function in a way similar to the human brain. DL is particularly powerful for solving very large datasets, which are common in many healthcare applications.

Turbo-charging drug development

Within drug discovery, the high failure rate strongly increases the average cost for development of a new drug. Biotech companies are trying different approaches to leveraging AI in drug discovery.

For instance, the company Atomwise is creating AtomNet, a deep-learning neural network based on molecular structures. AtomNet is designed to predict, in a high-throughput manner, the bioactivity of small molecules. This prediction can be used to discover new hits, find molecules with optimized selectivity, predict off-target toxicity or repurpose existing drugs.

Many other companies, including Recursion Pharmaceuticals, Exscientia, Deep Genomics, Benevolent AI, TowXAR, Numedii and Numerate, have also secured funding to explore the field and apply their AI algorithms in drug discovery and development. These companies all claim that they can significantly cut the cost of drug development and reduce the time from drug discovery to launch.

“It is too early to predict the exact impact of AI-based drug discovery on the overall development time and cost. Even though the early results are very exciting, we believe AI can be even more impactful by increasing the efficiency of the clinical development,” says Tarek Roustom, MD and Junior Analyst at V-Bio Ventures.

Leveraging clinical data

The mission becomes much more challenging, and rewarding, when employing AI in clinical decision-making or predicting clinical trial outcomes. The difficulty here is that the input data is usually very complex and hard to obtain.

As an example of a company active in this field, Flatiron Health is developing a learning healthcare platform in the form of a cloud-based database that includes medical records and clinical data from millions of cancer patients. AI-based analysis of past medical records allows to predict outcomes and propose optimized treatments for newly diagnosed cancer patients. That information can then be made accessible to clinicians, researchers, universities and drug-development companies. Flatiron raised over $300 M from different investors, including Google Ventures.

“The more we train these clinical database algorithms, the better they get. That’s why you see major funding flowing in AI-leveraged clinical data start-ups like Flatiron Health and iCarbonX. There is a big need to overcome the scarcity and variability of real-life clinical data,” comments Roustom.

No more eyeballing at scans

One obvious application for AI is using machine-learning algorithms to process large sets of unstructured data as machines are able to objectively spot complex patterns and correlations that humans wouldn’t come up with easily.

The low-hanging fruit here is the use of the ML and DL algorithms on data from different omics or imaging analyses, which are relatively easy to obtain. The resulting patterns can be very valuable to diagnose diseases, predict treatment response or unveil new therapeutic targets.” – Roustom

For example, Enlitic applies deep learning in radiology image recognition. Their technology can interpret a medical image in milliseconds, whereas a radiologist needs a couple of minutes. In addition, in June 2016, the Economist reported that in a test against three expert human radiologists working together, Enlitic’s system was 50% better at classifying malignant tumors. Moreover, Enlitic’s AI technology had a false-negative rate (where a cancer is missed) of zero, compared to 7% for human diagnosis.

Many other companies are using ML to discover patterns in biomarkers or imaging devices to diagnose different diseases, including Arterys, Butterfly Network, 3scan and DNAlytics.

Telemedicine and remote patient monitoring

AI platforms are being developed to provide patients with effective and low-hurdle health monitoring and advice. Such platforms can either complement or replace part of the work currently done by healthcare professionals, such as regular health assessment and monitoring.

The UK-based Babylon Health, for instance, runs a digital platform inviting patients to make virtual appointments with GPs and specialists. Babylon is also developing AI algorithms to chat with those patients and deliver simple health advice.

Other examples of companies developing patient-oriented AI platforms are Sense.ly, AiCure, SkinVision and Sentrian. Only the future will tell which of these will succeed in building large-scale customer engagement.

What will the future bring?

Roustom is excited about all the possibilities AI can bring to the biotech and healthcare sector in the broad sense: “We are optimistic that with the continuous advancement and the growing availability of affordable ML and DL service providers, the AI-applications will be increasingly adopted across the healthcare sector. AI will probably become a staple in healthcare, assisting researchers and clinicians in everyday tasks and hopefully creating shortcuts in drug development and disease diagnosis. We further believe that any successful AI-leveraged biotech start-up will have to be a biotech company at its core as AI will mainly be a tool to do some tasks faster and smarter. AI alone won’t create successful biotechs, but it is a valuable aid that an increasing number of biotechs will use. We expect that some AI-leveraged biotechs that fail to develop effective healthcare-related products per se will turn to the business of providing AI services.” For example, the AI discovery company Numerate, seeded by Atlas, pivoted to a service and collaboration business model. According to Roustom, these service activities will have to be highly focused on a particular niche: “Biotech-specialized AI service companies will face competition from tech companies providing generic AI services, including the likes of IBM and Google Cloud, that offer very affordable ML services.”