Sofia Tomov, an 18-year-old Bulgarian-American prodigy has founded Qardian Labs. She says “We develop innovative AI-based software solutions for assessing heart disease risk and are currently working with the US Air Force to develop an echocardiogram analysis system for decision support. Our software uses deep neural networks to achieve 99 percent accuracy on test data. The software also benefits sports medicine physicians by helping athletes pinpoint the risk of sudden cardiac arrest, the leading cause of death in athletes.
She is a winner of first place in medicine at the 2018 Intel ISEF for her research that uses machine learning to improve the diagnosis of heart disease. She reports: “As an aspiring computer scientist, I have been working on projects related to algorithms for genomic analysis as well as machine learning. In 2017, I won first place in the international Project Paradigm Challenge competition and was a finalist in the Discovery Education/3M Young Scientist Challenge. I was recognized by Business Insider as one of the “15 young geniuses already changing the world” and my work has been featured in US News and World Report and the Telegraph (UK).
Community activism is also important to me, as I have founded Teen Vote, a non-profit organization dedicated to lowering the voting age and promoting civic education, started a local chapter of Project CS Girls to teach middle and high school girl’s computer science, and volunteered teaching science and engineering to elementary school students.”
Sofia Tomov is:
- Passionate about solving problems and helping others.
- She is a student at the University of Tennessee at Knoxville. “As a computer scientist, I have worked on algorithms for genomic analysis as well as machine learning.”
- First place winner of Intel ISEF 2018 (Translational Medicine)
- First place winner of the international competition Project Paradigm Challenge 2017 (ages 9-13)
- Caroline D. Bradley Scholar 2017
- Finalist in the Discovery Education/3M Young Scientist Challenge 2016.
- Recognized by Business Insider as one of “15 young geniuses already changing the world.”
- Her work has been featured in US News and World Report and the Telegraph (UK).
- She founded Qardian Labs, a company based on the heart disease diagnostic software she developed and won first place at Intel ISEF.
- Founded Teen Vote, a non-profit organization dedicated to lowering the voting age and promoting civic education and volunteer teaching of science and engineering to elementary school students.
- Founded a local chapter of Project CS Girls, an organization dedicated to empowering middle and high school girls to change the world with computer science
The algorithm for the side effects of prescription drugs
Make the pill fit the sick person
Side effects of prescription drugs are the 4th leading cause of death in the US, according to the AMA.
“I wondered why people respond differently to the same drugs and found that genetic mutations affect a person’s response. However, knowing the mutation only solves half the problem. Doctors need to know if their patients have certain mutations. They need to identify a base change in a 6 billion base genome. To address this problem, I innovated a kind of computer algorithm, or problem-solving process, that can find mutations in a genome.”
Research in this area could lead to groundbreaking medical innovations that will alleviate pain by enabling personalised medicine, allowing doctors to create personalised treatments based on an individual’s unique genetic make-up.
How it works
Doctors can use string search algorithms to find mutations. String search algorithms are computer programs that find a string of letters in a text. The genome is the text and the mutation is the string.
The problem with most existing string search algorithms is that they are very slow. Speed is crucial in emergency situations, such as seizure or allergic shock, when the right medicine needs to be prescribed immediately.
“To address this problem, I implemented the Reverse Factor algorithm and parallelized or coded it to run on multiple computer processors. This has not been done for this algorithm. How does parallelization make an algorithm faster? A parallel program splits the task of searching for a genome by having multiple processors work together on a problem. This reduces the time required to find the solution. The parallelization of the Reverse Factor algorithm sped up the original by 400%, suggesting its feasibility. My results suggest that a string search algorithm like mine can save lives by advancing the field of personalized medicine and allowing doctors to predict a patient’s response.”
Heart disease is the leading cause of death worldwide, killing 20 million people annually. An accurate and timely diagnosis could be the difference between life and death for people with heart disease.
I used machine learning to improve the accuracy of diagnosing heart disease. Machine learning is a type of artificial intelligence that teaches the computer to learn from existing data and predict outcomes for future data.
For my work, I innovated HEARO – Heart Evaluation for Algorithmic Risk-reduction and Optimization. HEARO teaches the computer to act as an artificial brain using variable layer deep neural networks with normalization for data classification and classification. It is a unique contribution to the field in two important ways: 1) it surpasses existing results, including those from Stanford, and 2) the combination of a deep neural network and normalization is a new diagnostic tool. It has the potential to give doctors a potentially life-saving tool to make more informed diagnoses and allow people to receive treatment when they need it.
With the wind
Pollution from burning fossil fuels is a global problem. It is responsible for 1 in 8 deaths worldwide, according to the WHO. A cleaner source of energy, such as wind power, will save lives.
One obstacle to the use of wind energy is the inaccuracy of wind turbine power generation forecasts. Inaccurate forecasts force utilities to use backup units. This results in wasted energy. Inaccurate forecasts can cause utilities to waste money because backup plants are expensive to maintain.
I used machine learning to improve the accuracy of turbine power predictions. Machine learning is a type of artificial intelligence that teaches a computer to learn from existing data and predict outcomes for future data.
For my project, I innovated with HAWC – Hybrid algorithms for wind energy calculation. HAWC teaches the computer to act as an artificial brain using neural networks to classify and sort data.
HAWC can reduce the need for wasteful backup units, making wind power more feasible and reducing pollution.
How it works
The HAWC software set consists of the following algorithms: linear regression, polynomial regression, 2-level neural network and 3-level neural network.
Initially, I coded linear regression and polynomial regression in MATLAB programming language. Linear regression uses a linear function to make a prediction, creating a line that fits the data.
Polynomial regression uses a polynomial function to create a curved line. Each point represents a prediction, so an accurate fit represents an accurate prediction.
Part of the challenge of coding a neural network is structuring it to be accurate and efficient. It must be determined how many nodes or artificial neurons will be used and how many layers or steps to create connections will be used. “To address these challenges, I coded a 3-layer neural network that has not been previously applied to wind energy forecasting. Pollution kills 7 million people a year – that’s 13 people every minute. In the time it may have taken you to read this, pollution killed 65 people. Time is running out to find a solution to this crisis. My results suggest that HAWC could help reduce pollution and save lives by enabling wind energy development.”
by Vicky Bafataki Translation: Panagiotis Andreadis