User:Birulik/Insilico Medicine

Insilico Medicine is an AI-focused biotechnology company, specializing in the application of deep learning, generative adversarial networks, and reinforcement learning for biological target discovery, biomarker development, drug discovery, analysis of clinical trials and other fields of digital medicine. The company’s main focus is on dermatological diseases, programs in oncology, fibrosis, Parkinson's disease, Alzheimer's disease, ALS, diabetes, sarcopenia and aging in general. The company is headquartered in Hong Kong and operates 6 research and development facilities worldwide.

History
Insilico Medicine was founded by a computer scientist and biotechnologist Alex Zhavoronkov in 2014. The company started work in the field of drug discovery by using deep neural networks to identify potential drug targets by mining biological data. Insilico Medicine's research focus evolved in 2015 when Zhavoronkov applied Ian Goodfellow's works on machine learning, employing generative adversarial networks (GANs) in order to devise molecules with the desired medicinal chemistry properties de novo instead of screening existing databases of chemical compounds with the first peer-reviewed publication in GANs for drug discovery published in 2016. Following initial research into this approach, Insilico pioneered the use of GANs and reinforcement learning in drug discovery, becoming increasingly attractive to investors from AI and biotechnology fields. Since its establishment the company has raised over $50 million from venture funds and private longevity-focused investors, such as British philanthropist and entrepreneur Jim Mellon. Insilico's largest $37 million Series B investment round in September 2019 was led by Qiming Venture Partners joined by a number of funds including Eight Roads Ventures, F-Prime Capital, Lilly Asia Ventures, Sinovation Ventures, Baidu Ventures, Pavilion Capital, BOLD Capital Partners, and Juvenescence.

Recognition
Insilico Medicine's AI-based approach to healthcare has brought it industry wide recognition. At the 2015 Palo Alto Personalized Medicine World Conference it was titled the most promising company in the same year the company became Nvidia’s Emerging Company Summit finalist. Furthermore in 2017 Nvidia listed Insilico Medicine among top 5 social impact AI companies and included Insilico into its Nvidia Inception program. Insilico also found itself on CB Insights list of top ten private anti-aging companies in 2017 as well as the "A.I. 100" rating of most promising AI companies in 2018. Finally, in 2018 the company was granted the Frost & Sullivan North American Technology Innovation Award for advances in aging research and drug development. MIT Technology Review included AI-discovered molecules as one of ten breakthrough technologies in its annual review and highlighted Insilico Medicine's advances and key role in that field. In 2020 the company was included in the Fierce 15 list of top biotechnology companies by Fierce Biotechnology.

Drug discovery
Insilico Medicine pioneered the use of generative adversarial networks (GANs) and reinforcement learning (RL) in the field of drug discovery. In contrary to traditional approaches, when medicinal chemists search for potential drug candidates in molecular libraries, Insilico's drug discovery engine is built upon AI imagination that devises novel molecules with intended properties based on past research and information about patented compounds with proven efficiency against specific biological targets.

In 2016 Insilico published its first proof-of-concept for the use of deep learning algorithms to predict therapeutic use of medicinal chemicals based on information about other compounds. . That year the company proposed the application of GANs, specifically the generative adversarial autoencoders in drug discovery. The team trained a form of GANs called the generative adversarial autoencoders (AAE) using molecular fingerprints of anti-cancer drugs to identify drug candidates in PubChem database. Insilico's DruGAN (Drug + GAN) model, presented in 2017, introduced the transition from binary fingerprints to novel representation of molecules based on molecular graphs, thus allowing it to propose new molecular structures instead of matching output to molecular libraries.

In 2019 Insilico Medicine and its research collaborators from the University of Toronto published a proof-of-concept paper on the use generative tensorial reinforcement learning (GENTRL) for drug discovery in Nature journal. The researchers trained the generative model on the datasets of known inhibitors of DDR1, a tyrosine kinase target involved in progression of fibrosis, and tasked it to develop novel compounds that target DDR1. In 21 days the AI generated a large number of molecules that were put through sorting, scoring and review by chemists, with the most promising compound successfully passing in vitro and in vivo tests during the following 25 days. The team found the molecule to be potent against DDR1 and have drug-like qualities. The publication of these results attracted attention and became 2nd most popular scientific paper in the Nature Index in September 2019. The experiment was generally considered substantial advance in generative drug discovery. This work was criticized by a team from Relay Therapeutics which pointed out that the only AI-generated compound tested in mice was similar to a known molecule.

In late January to February 2020 Insilico Medicine used its machine learning methods to generate molecules that can be used in treatment of SARS-CoV-2 coronavirus. The company's AI was tasked to seek potential inhibitors of 3C-like protease, an enzyme critical for coronavirus reproduction, and was able to propose new molecules in 4 days. Insilico then published the research materials including molecular designs on its own website and ResearchGate, providing open access for researchers to examine and critique. Insilico Medicine promised to synthesize and test the most promising compounds and announced the search for industry collaborators to develop the following compounds.

As of the beginning of 2020 Insilico Medicine also ran drug discovery programs for cancer, aging, fibrosis, Parkinson's disease, Alzheimer's disease, amyotrophic lateral sclerosis, diabetes and other conditions. One of the drug candidates developed by Insilico, the treatment for hair loss, is scheduled to start Phase 1 clinical trials.

Biomarkers of Aging and Disease
Insilico Medicine works on different biological aging clocks based on biomarkers in the blood, gut microbiome and other sources of data about an individual. The company uses deep learning models to find reliable predictors of biological age (correlated with health status and mortality, rather than chronological age which merely reflects the number of years one has lived) and to understand what healthy aging looks like. If validated, such models may be used by researchers to track response in future clinical trials of anti-aging treatments.

Insilico's initial research into "hematologic aging clocks" began in the period 2015 to 2016 when the company trained a modular system of 21 deep neural networks with 60,000 samples from common blood biochemistry and cell count tests of relatively healthy individuals. The samples were linked to chronological age and sex. The researchers found the system capable of estimating a person's age within a time frame of 10 years with 83,5% accuracy and of determining a person's sex with 99% accuracy without measuring hormone levels. The paper was published in Aging and became the second most popular publication in the journal's history. That model was also used to power Aging.AI, an online platform that allowed individuals to determine their bioloigical age and sex by inputting their blood biochemistry data. Future research by the firm included deep learning analysis of poplulation-specific samples belonging to Canadians, South Koreans and Eastern Europeans allowing Insilico to improve precision within a margin of 6 years. Insilico Medicine also used its age-prediction model to analyze difference between the biochemical markers of smokers and non-smokers and quantify the acceleration of biological aging due to tobacco consumption, proposing a reliable deep learned method to determine an individual's smoking status.

In 2019 Insilico Medicine proposed a "microbiome aging clock" based on deep learning analysis of an individual's gut microbiome, which is known to change throughout adulthood. Researchers gathered information about microbiomes of 1,165 Europeans, Asians and North Americans with roughly one-third of samples coming from populations in their 20s to 30s, 40s to 50s, and 60s to 90s respectively. Ninety percent of samples were tagged with age in order to train deep neural networks that were later tasked to guess the age of persons among the remaining ten percent of those sampled. As a result, Insilico's algorithms were able to predict an individual's age within a margin of 4 years. Insilico Medicine then suggested that this aging clock, if validated, could be combined with other biomarker predictors to provide a more precise picture of individual's health and biological age.

Tools

 * MOSES (acronym for molecular sets) is the benchmarking platform, launched by Insilico Medicine to facilitate collaboration in AI-driven drug discovery. It contains an unified molecular dataset based on ZINC database, several molecular generation models by Insilico, Neuromation and Alán Aspuru-Guzik's research group, and evaluation metrics. MOSES aims to become an academic standard for generative models evaluation, comparison and sharing in drug discovery, accelerating the AI development in the industry the same way ImageNet has significantly boosted accuracy in image recognition Insilico Medicine released the MOSES source code as open-source without restrictions or copyright ownership claims.


 * In silico Pathway Activation Network Decomposition Analysis (iPANDA) is the method used to analyze activity of signaling and metabolic pathways, related to different conditions and diseases, using gene expression data. It was developed by Insilico Medicine in collaboration with Johns Hopkins University, Albert Einstein College of Medicine, and Boston University, Novartis, Nestlé, and BioTime. iPANDA allows to identify biomarkers for personalized treatment. It also allows for biologically relevant dimensionality reduction in order to effectively train the deep neural networks, that will predict the efficacy of specific compounds in treatment.


 * Geroscope is Insilico Medicine's drug knowledge management system based on the data mining of compound databases, accessed through partnerships and collaborations. The algorithms aggregate huge amounts of genomic, transcriptomic, epigenetic, proteomic and clinical factors, as well as data on the effect the drugs play on gene expression in cells, cell lines and tissues to identify potential geroprotectors.

Partnerships and Associations
Insilico Medicine maintains over 150 academic and industry partnerships, focused on validation and application of its technologies in different areas, such as target identification, drug development, predictive analytics in biotech. The company also relies on its partners to run clinical studies. Its Notable industry collaborations included Pfizer, Jiangsu Chia Tai Fenghai Pharmaceutical, WiXu AppTec, other big pharmaceutical companies, TARA Biosystems, Beijing Tide Pharmaceutical, and others. Insilico's partners in academic field include the University of Copenhagen, Gachon University, and Gil Medical Center in South Korea, Oxford University, Ageing Research at King's (a research center at King's College London) in United Kingdom.

Insilico Medicine has formed two joint ventures with its strategic investor Juvenescence. The first one being Generait Pharmaceuticals (formerly named Juvenescence.AI), that works on clinical development of AI-generated molecules. Insilico's second venture is Napa Therapeutics, established together with Buck Institute for Research on Aging and focused on development of compounds targeting NAD$+$ metabolism in aging-related diseases based on research by Eric M. Verdin. Consortium.AI, the company established through a venture by Insilico and A2A Pharmaceuticals, focuses on development of highly selective treatment for duchenne muscular dystrophy using AI-powered drug discovery engine. Another joint venture named Longenesis was founded by Insilico and blockchain mining and management company Bitfury Group to develop a blockchain solution for healthcare data management.

In 2018 Insilico Medicine, as well as major pharmaceutical companies, technology developers and research organisations working on AI application in healthcare announced the creation of an industry-wide collaborative coalition named Alliance for Artificial Intelligence in Healthcare (AAIH). The list of founding members included, but isn't limited to AWS, Bayer, GE Healthcare, GlaxoSmithKline, Johnson & Johnson, as well as University of Pittsburgh. Its first board meeting and official launch took place in San Francisco in January 2019, followed by an official statement by Alex Zhavoronkov. The Alliance aimed to increase education outreach on AI, promote investments in AI R&D and develop policies and regulation in cooperation with governmental bodies in EU, US and beyond. In October 2019 AIIH presented its first white paper "Artificial Intelligence in Healthcare: A Technical Primer" aimed to introduce concepts, standards and the potential of AI to broader healthcare community.

Locations and management
Insilico Medicine is headquartered in Hong Kong and employs 85 AI experts and scientist in research and development facilities in Belgium, Russia, South Korea, Taiwan, UK and the US. The company's recruitment strategy is focused on hackathons and competitions. For example, the company hosted Molhack II online hackathon in bioinformatics, biochemistry to support the establishment of its office in Taipei. Insilico's founder and CEO is Alex Zhavoronkov. Insilico is active in research publications and collaborations: as of 2019 its research team published over 120 peer-reviewed papers with over 3300 citations.