In 2019, the drug discovery process took ~15 yearsOpens a new window and cost several billion dollars.
Today, it takes four to five years, on average, to search for a new drug candidate.
Innovative companies produce first-in-class drug compounds within seven months.
Artificial intelligence solves drug-discovery problems in much less time and at a much lower cost.
While some of the world’s brightest minds work in this industry, it doesn’t take a PhD in quantum science and engineering to predict that AI would make a fine lab partner. That’s a no-brainer. But even the most intelligent people have limitations, so they invented AI to to handle abundant data efficiently and extract information pertinent to projects and pipelines.
Why AI for Drug Discovery?
In short, AI produces better results faster. The technology allows researchers to discover new drug candidates, advance treatment of diseases from cancer to Crohn’s, and reap the advantage of scale without the false-start drawbacks associated with traditional research and development methods.
AI can provide new answers in biology and chemistry that humans have not yet considered. Standigm, for example, has novel target identification (biology) and lead generation (chemistry) AI – proprietary platforms that generate commercially valuable drug pipelines. By contrast, humans cannot keep up with the rapidly increasing scientific literature. Manual information gathering and expert-only knowledge can only take biotech so far. AI can overcome many more constraints and challenges in the drug discovery industry.
Molecule Modeling and Prediction
Target identification is a big challenge in therapeutics. Although most AI-based drug discovery companies focus on creating new compounds and re-creating new drugs, only a few companies have the technology to discover new targets. AI streamlines target research. Modeling provides evidence for prediction and significantly improves the success rate.
Based on the vast biomedical information known to date, deep-learning knowledge graph analysis can align specific diseases with the proteins capable of targeting them. AI drug discovery companies are developing methodologies for scoring and ranking protein candidates and disease candidates expected to target specific proteins. Random-walk models apply RNA profile data. Metabolic models incorporate genetic information. Meta-learning-based drug-target binding predictors produce higher prediction accuracy and outperform baseline models.
Modern novel-compound lead design outshines traditional methods by applying advanced algorithms – Monet, NoSH, ChemMap, and IDEA-STAR for deep-learning-based structure/substructure optimization, scaffold hopping, and active learning – to generate patentable lead compounds for given targets.
Molecule modeling, including efficient models to encode graphs, is a key challenge of molecular representation learning. Transformer is a natural choice for graph processing but requires explicit incorporation of positional information. Existing approaches either linearize a graph to encode absolution position in the sequence of nodes – which loses the precision of relative position from linearization – or use bias terms to encode a relative position with another node – which loses the tight integration of node-edge and node-spatial information. Relative positional encoding, however, lacks these weaknesses. Methods to encode a graph without linearization, considering both node-spatial relation and node-edge relation, are more successful.
More sophisticated modeling and simulation methods are making more accurate predictions, including receiver conformer generation, ligand pose prediction, enviable docking, and free energy permission. Scientists then work to optimize the structure of lead compounds concerning the desired chemical properties, such as biological activities and ADME/Tox properties. Researchers can use this information to select the best molecule for further investigation. Down the line, they can determine whether making a compound would be easy or difficult.
Performing this advanced modeling first will eliminate a lot of trial-and-error experimentation later. Rather than doubling down on the time-consuming, people-intensive, physical experiments, biotech companies are pouring efforts into improving AI/machine learning (ML) approaches.
Automated Molecular Design Workflow AI
Other AI drug discovery trends include platforms that generate new drug candidates for a given target, and platforms that mine data from already-approved drugs for drug repurposing. What sets modern, innovative drug discovery endeavors above and apart from competitors is a more advanced iteration of artificial intelligence called Workflow AI, an automated molecular design workflow that can cover the entire drug discovery process.
Workflow AI will one day complete a closed-loop system of interconnected AI platforms whereby any data produced in any project is fed back for continuous improvement. This type of machine learning will reduce the resources necessary to secure patentable lead compounds inhibiting a novel drug target.
See More: How AI and Computer Vision Shape Our World
Competition vs. Shared Experience
Competition for drug and AI algorithm patents may be fierce, and production of platforms may be proprietary, yet still some of the best resources out there are free and readily available to researchers anywhere. A public user interface that essentially hands anyone the keys to one of the core functions of modern drug discovery. Scientists can enter the interactive environment – an AI-based target-identification platform with natural language processing technology combined with a pharmaceutical research platform – to explore, experiment, and discover new targets. The interface offers to prioritize protein targets for a given disease, then provides results on a graph with evidence of key paths among the disease-target connections.
The open interface offers a simple search bar to collect information from thousands of pages of scientific literature as well as numerous biomedical databases. It extracts the latest updates on gene-disease association from textual data, like scientific publication abstracts. Users can also check out a customized database where scientists and developers in the research ecosystem (pharmaceutical industry, hospitals, and academic institutions) can add their data to potentially uncover new targets. The interface has encouraged drug discovery research in specialized fields, including rare-disease areas.
The Ultimate Goal in AI Drug Discovery
AI drug discovery companies are quickly partnering, both publicly and privately, with pharmaceutical giants, biopharma companies, universities, and research institutes. Together, their pipelines are producing strong opportunities for treatments for diseases with high unmet medical needs, such as Parkinson’s disease, Alzheimer’s disease, Crohn’s disease, mitochondrial disease, cancer, rheumatoid arthritis, non-alcoholic liver disease (NASH), conditions such as autism spectrum disorder, and the list goes on. The collaborations will change the medical world and the world at large.
How do you think workflow AI drug discovery can fast-track Target identification and lead generation? Let us know on FacebookOpens a new window , TwitterOpens a new window , and LinkedInOpens a new window . We’d love to hear from you!