Inspiration

Living in Canada they all had direct access to affordable health care, but from a young age, each individually realized most people in the world don’t have this same luxury. By travelling to countries that didn't have the same opportunities as Canada, they all were quickly exposed to the degree of poverty in the world. More shockingly, the little access to healthcare came as a surprise to all of them. At home, the immediate access to healthcare did not feel like a luxury but instead a basic right, and this difference between countries sparked interest in each of them about the health industry. Years later after these realizations, the founders would end up meeting and discover their shared curiosity in the medical landscape. They all wanted to bring health care and affordable medication to everyone in even the poorest countries but were unsure of how to do so. After doing research they discovered shocking facts about the drug discovery process. Drug discovery takes on average between 12 and 15 years and has $2.6 billion in costs associated with it. More significantly, 90% of diseases don’t even have any type of treatment due to their rarity, leaving patients helpless. After understanding root causes in the drug discovery pipeline that prevented people from getting treatments they needed, Sigil, Aryan, and Will wanted to use their shared knowledge of artificial intelligence to solve these issues. After doing more research on how AI was already being used for drug discovery, they created Synbiolic, which took a unique approach to drug discovery.

What it does

Synbiolic is an AI-powered end-to-end rational drug design platform with a mission of making medicine more accessible, worldwide. Researchers are able to choose their targets, the small molecule’s property they wish to optimize and get a tailored list of small molecules that could work as a drug, as well as instructions on how to synthesize that molecule powered several of our custom state of the art machine learning technologies.

How I built it

Synbiolic is built using Amazon Web Services. We've built several proprietary state of the art machine learning models and deployed our models via AWS. Our models are trained on several molecular datasets totaling 4 million molecules.

Synbiolic's AI drug design pipeline has three core technologies that leverage a combination of machine learning algorithms developed in house such as deep probabilistic networks and reinforcement learning. Our three core technologies include our generative technology, molecular property technology, and retrosynthesis technology.

  1. Generative Technology - Synbiolic's generative technology is responsible for generating novel small molecule therapies that bind against a user-specified disease-causing target protein in order to find potential treatments for that disease. Synbiolic is able to generate molecules that can effectively bind against the target protein while also demonstrating the desired property 92% of the time which is unprecedented. Machine learning methods used to generate novel molecules often fail to accurately control the chemical property of the molecule so that it binds to a specific protein and demonstrates the desired property; however, Synbiolic is able to do so with extremely high accuracy.
  2. Molecular Property Technology - Synbiolic's molecular property technology takes the generated molecules in as input and evaluates the molecular property of the molecules. Synbiolic provides virtual molecular screening using our molecular property technology which can predict the molecule's ability to inhibit the target protein, its number of hydrogen bonds, its synthetic feasibility score, and more.
  3. Retrosynthesis Technology - Synbiolic's retrosynthesis technology predicts the retrosynthesis pathways of the molecules, which you can think of as the molecular lego instructions on how to synthesize the molecules. Our retrosynthesis technology aims to speed up the inefficient process of manually trying to figure out how to synthesize a molecule as one of the major criticisms of computationally generated molecules is that these molecules are oftentimes too difficult to synthesize making them commercially inviable.

Challenges I ran into

Along this journey of constructing the MVP, our team encountered various theoretical challenges. These challenges required our team to become familiar with topics covered in university-level chemistry and mathematics courses. However, through this experience, we were able to learn the fundamental concepts like organic chemistry, retrosynthesis, Markovnikov's principles, etc. which taught us the skill of learning and moving quickly in an agile environment, the importance of figuring things out independently when we faced with it, and the emphasis on maintaining team dynamic. We often found that without one of these pillars, our team values would be misaligned, which would slow our overall progress. However, through these quick iterations, we found that having weekly checkpoints allowed us to stay on the same page along with forecasting challenges or points to consider. Together, these intangible and tangible skills have been an opportunity to simulate the journey and difficulty of building a startup.

Accomplishments that I'm proud of

  • Runner Up for Imagine Cup America by Microsoft
  • Semi-Finalists for Diamond Challenge (Entrepreneurship Competition)

What I learned

One important lesson that we learned as a team was delivering results in an agile setting. Being able to move fast, build and test features in parallel, while constantly communicating with each other helped us push our product to launch a lot faster. Working through this agile process was difficult for our team and along this journey, we learned the process in dividing roles/ownership and strategy for approaching difficult tasks. Along this journey of constructing the MVP, our team encountered various theoretical challenges as well. These challenges required our team to become familiar with topics covered in university-level chemistry and mathematics courses. However, through this experience, we were able to learn the fundamental concepts like organic chemistry, retrosynthesis, Markovnikov's principles, etc. which taught us about the importance of figuring things out independently when we are faced with it, and the emphasis on maintaining team dynamic. We realized that team values, establishing a collective mission and staying aligned to it and having weekly checkpoints allowed us to stay on the same page along with forecasting challenges or points to consider. Together, these intangible and tangible skills have been an opportunity to simulate the journey and difficulty of building a startup.

What's next for Synbiolic

Our team at Synbiolic plans to further develop our MVP and launch a partnership program with research labs or AI/BioTech startups. We plan to further validate and further refine our models through these partnerships.

Due to the complexity of the scientific research industry, before Synbiolic can scale our platform - it’s crucial that we obtain further validation of our platform by partnering with research institutions or pharmaceutical companies to perform preclinical studies and publish the results in a journal. We would require additional financial resources to fund our planned preclinical study. As well, Synbiolic is currently working on generating a chemical library of molecules to find potential treatments for covid19 and we need additional resources to help us gather more data, advance our model and address our cloud computational needs.

In terms of future expansion, we plan to enhance model generalizability by performing extensive validation studies with potential partners and iterate from there. Other new features Synbiolic plans on introducing include multi-parameter optimization to our generative technology as currently, our generative platform is capable of single-property optimization of molecules. Another feature we plan to introduce is ligand pharmacokinetic prediction in order to provide a more comprehensive activity profile of generated molecules such as off-target side effects and expand our target market from preclinical drug discovery to include clinical study modeling as well.

Built With

Share this project:

Updates