2022.02.24

From Start to Phase 1 in 30 Months: AI-discovered and AI-designed Anti-fibrotic Drug Enters Phase I Clinical Trial

On the 24th of February, 2022, Insilico Medicine entered a new era of its evolution, transforming from an end-to-end AI-first drug discovery company into a clinical-stage AI-powered biotechnology company.

We are thrilled to announce that we have successfully completed a Phase 0 clinical study and entered a Phase I clinical trial with our first-in-class anti-fibrotic drug candidate for a novel target discovered using our artificial intelligence platform Pharma.AI™. The total time from target discovery program initiation to the start of Phase I took under 30 months, representing a new level in therapeutic asset development speed for the pharmaceutical industry.

On this same day one year ago, Insilico Medicine announced the nomination of a preclinical candidate in Idiopathic Pulmonary Fibrosis (IPF) for a novel antifibrotic target, both discovered using our artificial intelligence platform Pharma.AI™, in under 18 months. It was a precedent among any existing AI systems at that time to achieve simultaneous major success in target discovery and drug candidate generation for a broad indication, such as fibrosis, and a historical proof of concept for the ability of deep learning to link biology and chemistry in an integral workflow.

Around 9 months later, after successful results in preclinical studies, the company initiated the first-in-human (FiH) study in healthy volunteers to establish dose and basic safety for the discovered molecule. With the results of the FiH study having exceeded our expectations, today we announce the start of a Phase I clinical trial evaluating ISM001_055, an anti-fibrotic small molecule inhibitor generated by our AI-powered drug discovery platform for the treatment of idiopathic pulmonary fibrosis.

The entire drug discovery path from the initial concept to novel target discovery and preclinical drug candidate nomination took Insilico Medicine a small fraction of cost and time for a typical preclinical program which is estimated to be around $430 million out-of-pocket expenses and above $1 billion capitalized, and takes anywhere from three to six years to finalize.

The antifibrotic targets were picked using AI with two main criteria: the target must be an important regulator of pathways implicated in fibrosis and the target must be important in aging. The AI-powered target discovery tools now available in the PandaOmics ™ platform were used for target selection and prioritization. Once the antifibrotic targets were discovered and prioritized, Insilico utilized its Chemistry42™ engine. This extraordinary drug development sprint has proved to be possible owing to Insilico Medicine’s end-to-end Artificial Intelligence-powered platform Pharma.AI.

To the best of our knowledge, this achievement sets a historical precedent for a company to be able to bring to clinical trials an AI-discovered molecule based on an AI-discovered novel target -- rapidly and at a low cost.

A number of drug candidates for other indications were also designed using Pharma.AI — for Insilico's own pipeline (e.g. a recent drug candidate nomination for kidney fibrosis), and for the company's clients (e.g. a recent success story with immunotherapy drug candidate for Fosun Pharmaceuticals) — illustrating the system’s ability to replicate success and generalize to other therapeutic areas.

Bringing efficiency to drug discovery, and hope to IPF patients

Idiopathic Pulmonary Fibrosis is a broad medical condition that is limited to the lungs and primarily affects older adults. As the disease progresses, the health of the patient gradually deteriorates leading to potentially life-threatening pulmonary failure. Currently, very few therapies are available to patients, providing limited options to fight the disease.

To build the initial hypothesis, we used our target discovery module PandaOmics and trained it on a collection of omics and clinical datasets related to tissue Fibrosis and annotated by age and sex. PandaOmics then performed sophisticated gene and pathway scoring using a family of iPANDA algorithms previously published in Nature Communications, and came up with relevant targets via deep feature synthesis, causality inference, and de novo pathway reconstruction. The target novelty and disease association scoring was assessed by a natural language processing (NLP) engine, which analyzes data from millions of data files, including patents, research publications, grants, and databases of clinical trials. As a result, PandaOmics revealed 20 targets for validation, and one novel intracellular target was prioritized for further analysis.

Next, we applied Chemistry42 — our generative chemistry module for drug discovery. This module includes an ensemble of generative and scoring engines that can "imagine" molecules from scratch using cutting-edge deep learning technologies pioneered in 2015 by Insilico Medicine for pharmaceutical research applications. Chemistry42 creates drug-like molecular structures with appropriate physicochemical properties. In this case, Chemistry42 was used to design a library of small molecules that bind to the novel intracellular target revealed by PandaOmics.

The series of novel small molecules generated by Chemistry42 showed promising results on target inhibition. One particular hit, ISM001, demonstrated activity with nanomolar (nM) IC50 values. When optimizing ISM001, we managed to achieve increased solubility, good ADME properties, and a favorable CYP inhibition profile, while retaining nanomolar potency. Interestingly, the optimized compounds also showed nanomolar potency against nine other fibrosis-related targets.

In follow-up in vivo studies, the ISM001 series of molecules showed activity improving fibrosis in a Bleomycin-induced mouse lung fibrosis model, leading to further improvement in lung function. These compounds also demonstrated a good safety profile in a 14-day repeated mouse dose range-finding (DRF) study.

The best-performing molecule of the ISM001 series was nominated as a preclinical drug candidate in December 2020 for IND-enabling studies. A final version of the drug candidate, ISM001-055, demonstrated highly promising results in multiple preclinical studies including in vitro biological studies, pharmacokinetic and safety studies. The compound improved myofibroblast activation, a contributor to the development of fibrosis. ISM001-055's target is novel and has potential relevance in a broad range of fibrotic indications.

After completing the IND-enabling studies, and in order to better understand compound distribution, establish a dose, and characterize the safety profile of the drug in humans, we initiated an exploratory microdose trial (first-in-human trial) of ISM001_055 in November 2021, conducted in Australia in 8 healthy volunteers. The results exceeded expectations with the findings of a microdose of ISM001_055 with a favorable pharmacokinetic and safety profile of the drug in humans which successfully demonstrated clinical proof-of-concept.

Today we announce the start of a Phase I clinical trial evaluating ISM001_055, having dosed multiple volunteers with our AI-generated innovative small molecule inhibitor -- a potential treatment for idiopathic pulmonary fibrosis (IPF).

The Phase I clinical trial is a double-blind, placebo controlled, single and multiple ascending dose study to evaluate the safety, tolerability, and pharmacokinetic of ISM001_055. In the study, 80 healthy volunteers will be enrolled in 10 cohorts consisting of 5 single ascending dose and 5 multiple ascending dose cohorts. The primary endpoints are to determine maximum tolerated dose and establish dosage recommendations for future Phase II studies.

The whole pre-clinical development program, from hypothesis and novel pan-fibrotic target discovery to preclinical drug candidate generation, took just under 18 months to complete at a budget of around $2.6 million, and additional 12 months — to get the preclinical drug candidate through successful Phase 0 to Phase 1 clinical trial. This accomplishment is several orders of magnitude faster and cheaper as compared to a traditional drug discovery process. Even more importantly, with this milestone, and with other drug discovery successes, the team at Insilico Medicine demonstrated that the AI-driven approach to drug discovery is what we believe a new way to conduct pharmaceutical research -- the future of drug discovery.

What it takes to become a leader in the pharmaceutical AI race

Identifying a novel pan-fibrotic target, developing a drug candidate with an unprecedented mechanism of action, and bringing it to a Phase I clinical trial in under 30 months may seem like a drug-hunter’s dream come true, but the path to this early success has been long and thorny. Being a pioneer in the AI-driven transformation of the pharmaceutical industry is a challenging yet rewarding journey.

We founded Insilico Medicine in 2014 as a deep learning company from early days, inspired by a then rising recognition of deep learning technologies as a global game changer in many industries, after a series of widely publicized successes, including a famous ImageNet competition where a convolutional neural net AlexNet demonstrated outstanding performance in image identification. The same year, Google announced that their deep learning system managed to identify cats in YouTube videos without any prior knowledge of or instructions on identifying cats -- the model did it through an unsupervised learning process. Those two milestones, among many other achievements in the area of deep learning, created excellent momentum for the AI field, and catalyzed an exponential growth in both theoretical AI research and practical applications of AI in many industries.

In 2015, around the period of time when Deep Learning gained global recognition and rose in prominence outside the pharmaceutical space, we started evaluating generative adversarial networks (GANs) for drug discovery applications. GANs are a type of deep learning architecture that have one neural net inventing new stuff to match some predefined requirements (a generator), while another neural net is trying hard to prove the generator wrong. Both neural nets are tasked with learning until the generator ends up with the optimal result. GANs use low-dimensional formats like binary fingerprints, SMILES strings, graphs, and other light representations to generate molecules.

We described the concept of using an Adversarial Autoencoder (AAE) for the generation of novel molecules in our paper "The cornucopia of meaningful leads: Applying deep adversarial autoencoders for new molecule development in oncology," submitted for publication in Oncotarget in June 2016. This publication coincided closely with the publication of a similar idea by Alan Aspuru-Guzik’s team in their ArXiv paper "Automatic chemical design using a data-driven continuous representation of molecules." During this period, we began to collaborate and build a global community around generative chemistry.

Later we developed several improvements and new features to our GAN-based AI platform for drug design and started patenting our findings. In 2017 we built multiple working GAN models, including druGAN for fingerprints, ORGAN for SMILES, various recurrent neural networks (RNN) architectures with reinforcement learning and LSTM, agile temporal convolutional networks (ACTNs), and a reinforced adversarial neural computer (RANC). In 2018 we progressed towards building and validating a powerful deep generative model, generative tensorial reinforcement learning (GENTRL). GENTRL is a new AI system for drug discovery that dramatically accelerates the process of lead discovery from years to days. We made the code publicly available on GitHub to encourage a broader community of scientists to continue building on this work.

Eventually, we built an end-to-end AI platform Pharma.AI with three key components: a target discovery and multi-omics data analysis engine PandaOmics, a de novo molecular design engine Chemistry42, and a clinical trial outcomes prediction engine InClinico.

During the last several years, we have also been investing heavily in the synthesis and validation of the molecules suggested by our engine for various projects using a network of contract research organizations.

Conclusion

To help the pharmaceutical industry expand the drug discovery and development capabilities, we productized and made available for licensing our target discovery and generative chemistry system. We invite pharmaceutical companies to join the Pharma.AI platform community–by deploying the PandaOmics ™ and Chemistry42 ™ systems, and piloting the inClinico ™ clinical trial outcome prediction system.