Artificial intelligence (AI) is the ability of machines to perform certain tasks ״intelligently,” which is often based on the intelligence and expertise of humans. But what if humans lack the expertise to train the machines?
This is often the case in drug development, explained Noam Solomon, CEO and co-founder of Immunai.
“Fewer than 10% of drug candidates identified by human experts and validated pre-clinically will eventually receive FDA approval,” he told The Jerusalem Post. As such, “we must deliver not human-expert parity but a complete paradigm shift if we want to discover new therapeutics more reliably and accurately.”
“The next level of research in immunology, medicine, molecular biology and chemistry will be powered by coupling new types of high-resolution measurement technologies across a myriad of experiments and patients, stitched together into large datasets by vertically integrated engineering infrastructure and mined for novel and actionable therapeutic insights,” Solomon said.
“The next level of research in immunology, medicine, molecular biology and chemistry will be powered by coupling new types of high-resolution measurement technologies across a myriad of experiments and patients, stitched together into large datasets by vertically integrated engineering infrastructure and mined for novel and actionable therapeutic insights.”
That is what Immunai, the company that Solomon co-founded and for which he serves as CEO, is trying to do.
What is Immunai trying to do?
Founded four years ago, his 150-strong team includes scientists and engineers with expertise in a range of disciplines from engineering, computational biology and machine learning to immunology and drug discovery and development. Immunai is leveraging cutting edge, high-throughput technologies to measure the complex cellular interactions within the human immune system using various data types at a single-cell resolution.
“We believe that a higher resolution atlas of our immune system should contain the signals that govern why certain drugs work for some patients and not for others,” Solomon said. “Humans – even experts – are not well equipped to make sense of such high-dimensional big data, but machines are. This is the paradigm shift. We are relying on machines and big data to significantly augment human expertise. The way to do this is to build a vertically integrated solution.”
The term vertical integration stems from the economic industry and traditionally refers to a business strategy in which a company controls multiple stages of its production process and supply chain. Vertical integration, as Solomon uses it, refers to the development of transformative engineering platforms that pull together all of their components into one cohesive unit. This, he said, is the true secret sauce of next-gen platforms and it goes far beyond AI – though AI definitely has a role in such platforms.
Solomon said autonomous vehicles are a good example of the type of vertically integrated platform he is referring to. Their “stack” contains the vehicle itself, all its sensors, vast quantities of pre-mapped data and the computer vision and planning algorithms. When integrated all together you have a self-driving car.
“It is an end-to-end solution in the sense that a person can get into a car and get from one point to another with little to no involvement at all by the person,” Solomon described. “The more the driving of the vehicle is done autonomously, without interference or involvement of external systems, and without manual intervention, the more the vehicle is vertically integrated.”
“Vertical integration is just as important and promising for drug discovery and development,” he continued, “but it also faces a fundamental and more challenging difference from self-driving cars or medical imaging: lack of strong human expertise with which to train.”
To fill this gap, an explosion of new technologies, data and accompanying algorithms have been developed in the last five years. Immunai is at the forefront of this work.
“We have created a vertically integrated, engineering-driven platform to map the human immune system … and find actionable insights to support pre-clinical and clinical progression,” Solomon said.
At the heart of the platform is “AMICA” – the company’s annotated multiomic immune cell atlas. Omic is another word for a suite of biological molecules that translate into structure, function and dynamics of an organism. One such example is “genomics.” Another is “proteomics.” Immunai is measuring the immune system with multiple such omics.
This is the largest-of-its-kind clinical-genomic dataset that measures and maps the immune system at single-cell resolution, including thousands of different datasets, spanning hundreds of disease indications, human samples as well as different preclinical models.
Using AMICA, Immunai can apply a suite of computational models to collect, integrate and harmonize data generated from patient samples.
“We apply a suite of bioengineering technologies on the single-cell level to generate the highest quality data from patient samples,” Solomon explained. “For each cell that we measure, we measure thousands of properties, leading to over a terabyte of information per patient sample, and use our AI/ML technologies to generate insights from the data collected.”
In recent years, more companies are talking about causal inference, which aims to validate that the insights you are finding in the data are “real.” To this end, Immunai leverages functional genomics capabilities, including using CRISPR gene-editing technologies, to identify and validate both preclinical and clinical hypotheses generated from the data collected within AMICA.
And with the richness and granularity of the data in AMICA comes the machine learning (ML): The company’s unique ML pipeline, including “SystemMatch,” “SignatureMatch” and “IC-GRN” enables the platform to identify and refine immunological signatures that can be relevant for novel therapeutics and diagnostics.
“The goal of our vertically integrated platform is to provide an end-to-end offering that improves the entire process of drug development,” Solomon said. “Over time, it should inform better clinical trial decision making and optimization of novel therapeutic candidates earlier, on fewer samples, and with higher accuracy and specificity.”
He said that talking about building an engineering infrastructure that supports vertical integration is “less sexy” than talking about the magic of AI, but he believes that doing so “is key to unlocking and disrupting the biopharmaceuticals industry.”