Learn more about what artificial intelligence (AI) is, how to implement it into your business, and its benefits.
Over the past three decades, humankind has developed an extraordinary ability to generate data. We gather data from almost anywhere: from mobile phones to traffic sensors. Raw data is usually complex and highly dimensional. For example, heart rate data may include a person’s physical data, time and period the reading was taken, whether a person was active beforehand, and how long the subject was awake, just to name a few categories. Researchers nowadays are leaning more on smart computational tools to make sense of these data, and help them find insights. AI can screen large, compound databases, finding insights and deriving knowledge that can be helpful.
The global AI market has been increasing exponentially, and biotech is one of the fields that has benefitted the most from AI. Applications in biotech include drug discovery applications, drug design, predicting pharmacokinetic profiles of compounds and compound toxicity, understanding 3D protein folding expression, and predicting interactions between drugs and molecules. One major breakthrough in science was being able to generate 3D protein models given an amino acid sequence, done using AI.
Medtech is also benefiting from AI. For example, many AI methods nowadays scan MRI or ultrasound images and compare it with large data sets of known diagnoses to determine a diagnosis. Recent medical device innovations can gather constant readings of people’s health. AI can support humans in interpreting and extracting insight from these medical device readings.
Natural intelligence can be subjectively defined as a property of the mind that encompasses abilities such as the capabilities to reason, plan, solve problems, think abstractly, comprehend ideas, use language, and learn.
Artificial intelligence, thus, is the capability of a computer program to perform tasks or reasoning processes that we normally associate with human intelligence. We see applications of this in computerized chess games and non-player characters in video games that simulate human behavior.
Machine learning is a subset of artificial intelligence, which IBM defines as “a branch of artificial intelligence and computer science which focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy."
The machine learning process can be generalized as key features being extracted from images, and then a computer model creating categories to generalize these features. For example, you can show a program a picture of a cat and a picture of a dog, telling it which is which. The program begins making generalizations based on specific features - like having two ears. As the machine encounters more pictures of cats and dogs, it refines its generalizations, which then can be applied to an unknown image.
The machine learning cycle has 5 stages: hypothesis, collect data, training the algorithm, try it out, and improve the model.
The machine learning cycle starts with a hypothesis - something you want to predict. An example of a hypothesis is, “I can predict diabetes in patients.” Collecting data would entail gathering blood tests, biomarkers, and other metrics of patients with diabetic and health patients, and showing these data to a machine.
You would then train the algorithm by showing and telling the machine which samples were diabetes positive and which were negative, going through many different samples. After training the algorithm, you would test the model on a sample that you don’t say its property. Finally, you would evaluate how your model performed on the test case, and then reiterate through the steps.
There are two basic types of machine learning methods: supervised and unsupervised. Supervised machine learning is the most common type of machine learning, where a person must tell the program what the output should be. After going through many examples, one can test if the program can recognize and classify data without being told the answer. In unsupervised learning, the computer program will simply receive all the data and begin organizing the data, and find its own patterns and classifications.
Deep learning is a subset of machine learning that is modeled after how humans process information. A computer scientist builds code called an “artificial neural network,” which is the code that simulates a human brain.
One of the biggest breakthroughs of AI in biotech was defined in a journal article called “A Deep Learning Approach to Antibiotic Discovery.” In these experiments, the researchers used deep learning to identify new antibacterial compounds that differ from existing antibiotics. In an age where antibiotic resistance is becoming an increasing concern, this methodology for identifying and selecting new antibiotics would significantly decrease the amount of time for finding new therapeutics.
The applications of AI fall into marketing recommendations, sales, human resource allocation, and many areas of supply chain optimization, and most of all, research & development (R&D). In R&D, researchers often have heaping data sets to sort through - a computerized model doing preliminary sorting gives researchers to work on more impactful tasks. AI is now an increasing factor for governments, investors, and innovators.
1. Take advantage of external and public data sets to train algorithms
Sometimes, one company may have a limited amount of data points gathered in-house. Companies can boost the capabilities of their algorithms by implementing and integrating this data with external data. It could be beneficial to create partnerships to share data with organizations known for collecting data that your company is processing. You can combine data longitudinally, for example by combining your patient records with the historical medical records of those patients to predict the outcome of treatments. You can also find cross-sectional data to combine your data with - such as the treatment cost of a certain disease. Looking at healthcare economics data can help you predict cheaper alternatives for a specific type of treatment.
2. Use AI to target bottlenecks in your business processes
You can use AI to address areas that are limiting the growth of your business. Once you identify what is holding back your business, such as fragmented data or human resource allocation, you can find a way to automate those tasks and expand the capabilities of your company. The most important step of this process is recognizing what is holding your business back. Once the problem is identified, implementing automation is a streamlined process that computer scientists can spearhead
3. Use AI to benefit, not displace, your workplace
While AI is a powerful tool to aid researchers and administrators, it is not going to replace human intelligence.
“The capability of processing enormous amounts of data goes beyond human possibilities…In any case, at the end of these processes, it is always human responsibility to evaluate the insights and… make informed and wiser decisions based on them.”4. Start small
When first implementing AI, do not try to solve all the problems you’ve pinpointed at once. Think big, but start small by staking a piece of your problem and implementing AI. Evaluate the implementation of AI in measurable metrics that you can show to others. Once you have implemented AI at a small scale with success, you can scale the implementation to solve other problems.
🔬 Related: What Is the Lean Startup Methodology?5. Utilize experts
You don’t need to know everything on your own. There are many existing AI consulting firms that can help businesses and startups implement AI into their business. This takes the burden off of the internal team, and the external consultants can bring new technologies, methods, and views to the issues your team is facing. That way, you will get a much better idea of the extent of AI for your business.
It is also possible to outsource parts of your process. If you are planning on outsourcing your processes to an external AI firm, it is important to talk with people that understand your business’s core values and scientific endeavors. If they don’t understand your business’s processes well, they may not be able to translate them into beneficial AI solutions, making the outsourcing worthless.
How does one manage projects involving AI with multidisciplinary teams?
It is important that everyone understands the goals and benefits of implementing AI. In the beginning stages when the team is designing experiments, teams must align expectations of what AI can bring as a solution. In some projects, the domain expert can be a medical doctor, so they would need to communicate clearly with the AI engineers what they need from an AI program - conversely, the AI engineers must communicate with the domain experts about the limitations and best practices of using AI.
How do you deal with projects that only have a little data or a lack of data?
First and foremost, you need to understand what kind of predictions can be made with AI. Also check if the quality and distribution of the data; having high quality and diverse data sets allows an AI to pull more insights from data.
You can apply methods that do well with small data sets, like decision trees or random forests. Sometimes, you may need to enrich your data by importing external data. There are also companies that specialize in generating synthetic data to enrich data sets.
What about other industries like banking?
The AI algorithms used across industries are often the same - for example, models that make financial projections are also used in climate forecasts. What makes the difference is the domain expertise, as different domain experts know where to apply AI to get maximum benefits. Knowing the subject expertise is important to understand the limitations of AI in the domain.
This article comes from a webinar hosted by University Lab Partners. The event speaker is Dr. Bruno Horta, co-founder and CEO of Join Analytics.
📽️ Watch the full webinar here.
Download The Ultimate Guide to Wet Lab Incubators in Southern California, a handbook to assist life science start-ups through the entire decision-making process to find wet lab space.DOWNLOAD NOW