top of page

Artificial Intelligence in Pharmaceutical Drug Development

Pharmaceutical formulation is a lengthy process that includes early drug discovery, preclinical studies, clinical development, FDA review, post-market management, and everything in between. Because these processes are lengthy and expensive, AI has rapidly become a highly useful tool in the scientific community. Many utilize AI’s extensive capabilities that range from basic, mundane tasks, to more challenging ones. This technology is capable of completing simple tasks such as mapping thousands of target documents with record speed, quickly sifting through all relevant scientific literature, and connecting data to one another. Furthermore, AI is also capable of more advanced applications such as designing molecules, identifying new drug targets and even designing experimental and clinical designs. Regardless of how the information is used, AI could be the future of scientific data collection and implementation.


There are numerous AI algorithms available, depending on how the extracted data is to be employed. In essence, AI algorithms are capable of using processed data and experiences to make predictions and recommendations. The technology can then assess and adjust the rules and algorithms as it sifts through more content and begins learning. AI algorithms notably interpret language the same way humans do via natural language processing (LNP) techniques. LNP is important for AI to uniquely adapt algorithms in response to new data exposure.


There are many things to consider when choosing a specific AI platform. The most important factor includes choosing a machine learning algorithm. There are four main types of algorithms, which include supervised learning, semi-supervised learning, unsupervised learning, and reinforcement learning.

  1. Supervised learning is allowed to learn by example, where the algorithm is given a known dataset that contains desired inputs and outputs. Corrections can be made to output predictions until the algorithm has a sufficient level of accuracy and performance. Supervised learning allows for classification, regression, and forecasting.

  2. Semi-supervised learning is similar to supervised learning, but can additionally use labeled and unlabeled data. This algorithm can begin to label unlabeled data by combining data that contains meaningful data with data that does not. Semi-supervised learning is also capable of classification, regression, and forecasting.

  3. Unsupervised learning identifies patterns in data, establishes correlations and relationships through accessible data analysis. Unsupervised learning includes clustering and dimension reduction.

  4. Reinforcement learning algorithms are given a set of actions, parameters, and end values for machine learning by trial and error. This allows for algorithm adaption to approach specific situations.

Another aspect to consider when developing AI is workload reduction and machine learning times. Workload reduction is an important application so that employees can focus on other aspects of a designated operation. To accomplish this, machine learning algorithms should include human-readable output which includes guidance on output interpretation and how employees can verify and respond. Workload reduction must also be time-efficient. Machine learning times should have a relatively quick turnaround. Understanding the time it takes for algorithms to trigger detections in new environments and how long learning periods are determines platform choice.


The last two considerations when choosing an AI platform include machine learning algorithm volume and the time it takes the platform to integrate with other systems. For example, Python is a useful and general language capable of completing various developmental tasks, which include linear and logistic regression, building decision trees, naive bayes, and much more. Using Python and many other languages allows for easy integration with other platforms. It is essential to choose platforms that have the ability to provide intelligence to an existing infrastructure to reduce response time. This integration is even possible through APIs (Application Programming Interface).


In regard to scientific applications, AI technology can accomplish many tasks. Some AI can target research that simply requires pulling, ranking, and grouping relevant papers towards a research question, while others are more advanced. Using this type of AI technology allows programs to sift through millions of papers and determine which are most reputable and relevant to a specific hypothesis or research topic in a fraction of the time an individual can. A few AI programs include Iris.ai, SourceData, Google Scholar, and many others.


Transitioning to AI’s role in therapeutics, there are many applications which include anything from sorting publications of interest to validating biomarker and drug-target discovery. More advanced algorithms are capable of relating pieces of data together, such as suggesting that two proteins interact with each other based on AI’s comprehension of their specific characteristics. It is also capable of target identification, molecular simulations, predict drug properties, de novo drug design, and develop synthesis pathway regeneration.


Iris.ai is one tool that specifically focuses on early phase research aid. This program specializes in forming a map around a specific research topic or personal statement. To do this, Iris.ai sorts through millions of publications based on extracted keywords, synonyms, and hypernyms to present publications worth reading. There is an additional Focus Tool offered that takes the initial compilation of publications and further sorts through which publications to include or exclude.


Other programs, Dimensions and Euretos, have advanced technology dedicated to research and development. Similar to programs such as Iris.ai, Dimensions and Euretos, can further pull and interpret articles in addition to understanding topics pulled from ontologies. Specifically, Dimensions offers programs ranging from AI’s role in research evaluation and analysis improvements to aiding the pharmaceutical industry with biomarker identification. In addition, Euretos AI technology can search for a specific protein and further categorize publications into extra dimensions. For example, the platform can categorize genes associated with a protein of interest and suggest potential targets. Euretos is capable of ranking the number of publications based on genes associated with the target protein. Euretos can go one step further and produce a visual diagram of a subject-affiliate connection.


In addition to identifying target biomarkers, proteins, and more, AI is also capable of drug target identification and formulation. PandaOmics is a discovery platform that uses transcriptomic and proteomic data scores and other text data such as molecular data, publications, scientific literature, and meta-data repositories to predict specific target genes for diseases. PandaOmics takes into consideration Omics AI scores, KOL scores, Finance scores, and others to quickly and efficiently formulate a list with respect to specific research goals.


After targets have been identified, a program called Chemistry42 can develop molecules using generative AI. This AI technology is capable of de-novo drug design by using standard active learning platform capabilities and a reward setting dependent on API. Chemistry42 will not only develop molecular structure, but also includes predicted ADME-PK and kinase selectivity characteristics. The biotechnology company, Insilico Medicine, has recently utilized Chemistry42 and other AI platforms for drug generation and is now moving into Phase 2 clinical trials. Insilico Medicine allowed AI to find a target biological mechanism pertinent to idiopathic pulmonary disease and further develop a compound to act on the target.


Although there are many AI platforms focused on drug discovery and development, there are others primarily focused on patient profiling and treatment marketing. IPM.ai is one platform focused on improving patient care by studying the development, clinical study, and market therapeutics. IPM is capable of maximizing commercialization efforts, estimating disease prevalence, referral network mapping, and more. This platform allows for identification of undiagnosed and misdiagnosed patients and further discovering what diagnostics and treatments were completed. This information is beneficial in understanding how valuable a product concept is, understanding the market opportunity size, and predicting the likely demand for a given treatment.


AI is continually advancing and gaining attention from life-science investors interested in these opportunity areas. Newer investments are geared towards AI learning for data gathering, drug discovery and development, prediction analytics, process modeling, performance optimization and other areas. Many financial solution firms have predicted up to 50 new treatments throughout the next decade with a $50 billion market result based on current investments in AI technology. One investment example involves Nvidia investing $50 million in Recursion’s AI system which specializes in drug discovery. Recursion Pharmaceuticals gives drug making companies access to drug design and development datasets from their AI generation cloud platform. As the demand for AI research increases, so do the vast platform capabilities.


We have compiled a list of currently available and relevant AI platforms specializing in early research all the way to clinical trials. Anyone who is interested in utilizing AI for research of any kind can benefit from this list. Some of the research capabilities include early research, data collection and sorting, drug discovery and design, target and biomarker discovery, experimental design, and pre-clinical and clinical design. At the bottom of this page you will find an Excel file that you can download with this comprehensive list. If an AI platform performs in more than one research area, it is then noted in the ‘stage of research’ column on the Excel sheet. The AI platforms are presented in the following categories:

  1. Name of AI platform

  2. AI capabilities

  3. Machine learning algorithms

  4. Stage of research

  5. Link to AI platform website

Hopefully you find this useful as a reference for your research or general education.


References

AUST Business Solutions_AI Platforms for Drug Development
.xlsx
Download XLSX • 18KB

103 views0 comments
bottom of page