Writing a brilliant statement of purpose for AI/ML-focused grad programs takes more than a sophisticated essay structure. It requires conveying an explicit direction in your career goals. You can’t simply say: “Hey, I majored in Computer Science and I made great grades and I want to become an ML expert.” That’s not enough. You have to have a thorough, meticulous reason. You have to have purpose. When someone says “Why do you want to study AI?” you must explain the focused subniche where you’re going to make a difference – that is the key to your machine learning SOP.
Luckily, when I met Aaron, I didn’t have to ask this question. He was super eager to explain all the wild new things happening with “in silico design of drug development.”
You see, Aaron wasn’t just looking to get a master’s degree or PhD. He wasn’t just looking to specialize in AI. He was looking for a specific education that would let him contribute to breakthroughs in disease diagnosis, personalized treatment, clinical trial research, radiology, and more.
For Aaron, AI wasn’t the goal – it was the tool. He needed the tool to pursue bigger goals. And this is why he was admitted to the best CS programs in the country.
Let’s examine his SOP and his overall amazing candidacy. Seriously, Aaron is an all-time rock-star. His example shows what it takes to get into the best programs, but it will also reveal how we can improve our own grad school applications.
- B.S. from top-25 US university
- Major: Bioengineering
- Dean’s List GPA (good, but not perfect)
- Research Experience: 1 year undergrad, 1 year full-time in industry
- All rejections during previous application cycle
Accepted to the #1-ranked CS program in the US, with multiple admissions from Top-15 schools (according to CSRankings.org)
It’s interesting to note that Aaron was applying for his second cycle. One year earlier, when he was still an undergrad, Aaron wrote a dreaded “autobiography” essay and received all rejections. This time around, he took a far more nuanced approach to his SOP, and had wild success. He did have one year of amazing professional experience, but note that the “formative” moment in his introductory paragraph happened when he was still in school.
It’s also worth nothing that Aaron was changing academic fields – from Bioengineering to CS. When he realized he wanted to make this change toward the end of undergrad, he took extra classes to get a CS specialization. Then, after graduating, he continued taking part-time NLP and Algorithm classes while working as a bioengineer during the day.
After hundreds of hours and thousands of pounds of disposable lab supplies, the catalytic enzyme I worked on at Dr. Bruce Banner’s lab only saw incremental improvement. Yet, Marinus, a cutting-edge physics based protein-folding software that I learned to use in this lab, replicated my months of work in the blink of an eye. The first time I saw a protein fold and dock to a ligand on my computer screen, I was thrilled. Though I had originally come to Dr. Banner’s Protein Engineering Institute to learn more about in-vitro enzyme design, I found myself more and more captivated by the exciting possibilities offered by in-silico design.
This excitement resonated shortly after, when I read about DSP-1181, an OCD treatment. According to Nature, the average drug costs $2.6 billion to develop, yet DSP-1181 was developed five times faster than usual using Exscientia’s AI tool, significantly reducing its development cost. Innovations like these fascinate me. Along with drug discovery, AI will soon lead to breakthroughs in disease research and diagnosis, personalized treatment, clinical trial research, radiology, and more. I believe that my background in Bioengineering, combined with a specialization in Artificial Intelligence in the Stark University MS/PhD Program, will allow me to contribute to these breakthroughs in a meaningful way, making healthcare affordable and improving patient outcomes.
At Gotham University, I majored in Bioengineering with a specialization in Computer Science, providing me with a strong foundation in Biology, Physics, Chemistry, Multivariable Calculus, Differential Equations, Linear Algebra, C++, Python, Matlab, Data Structures and Algorithms, and Computer Networks. In my capstone project, I created a congenital clubfoot compliance sensor which collected data to aid Dr. Otto Octavius and his peers in their disease research. This project finished in first place within the department-wide symposium. Furthermore, since graduating, I have continued expanding my CS competencies by taking additional coursework in NLP with Python & Data Structures and Algorithms through Stanford Continuing Studies and UC Berkeley Extension school respectively. In one of these classes, I led the development of a Random Forest anti-microbial peptide identification application based on frog-skin-secretion data. With these experiences, I believe my foundation in CS is robust enough to tackle the greater challenges of graduate study at Stark University.
Additionally, I have a breadth of lab experience ranging from Narcolepsy and Retinal Prosthesis labs at Stanford to catalytic enzyme-mimic research at Professor Erik Selvig’s Lab at the University of Sydney. As mentioned previously, I also spent a semester at Empire University as a Research Assistant in Dr. Bruce Banner’s laboratory, where I learned to apply Marinus to the evolution of a catalytic enzyme, a de novo beta barrel.
Today, at ShieldCodex, a protein engineering company, I conduct research to evolve enzyme therapies for various diseases through High Throughput Screening. During the development of a specific gene therapy, we found that we lacked the equipment to detect certain Post-translational modifications (PTMs) in the wet lab. I took the initiative to address this issue and I created a proprietary N-linked glycosylation detection and Tyrosine sulfation and Thrombin cut site prediction application. My Long-Short Term Memory (LSTM) application was the first PTM detection tool at my company, and it had a higher accuracy (Matthews correlation coefficient) on the dataset than any other state-of-the-art application in this problem area. This application led to the development of a gene that was certifiably safer before insertion into the human body, and the application is now being deployed in the company-wide software package for projects beyond its original scope.
My experiences have shown me that there is much work to be done to integrate AI and Biology. During development of my PTM tool, I saw that the industry standard NLP-based approach to Tyrosine Sulfation detection has limited capacity to capture protein structure data. Additionally, Tyrosine Sulfation data was scarce. Similarly, in my capstone project at Gotham University, I found that a shortage of data had stunted congenital clubfoot research. To capture the nuances of biological data and address biological data scarcity, we require AI experts to create tailored algorithms and tools, rather than out-of-the-box solutions.
Thus, my research and work experiences have shown me that the overlap between Biology and Computer Science is ripe for exploration, and Stark University’s MS/PhD Computer Science Program seems perfectly suited to pursuing this goal. Specializing in Artificial Intelligence will allow me to formalize and further build ML expertise. I am excited by the prospect of coursework like CS311A: Machine Learning: From Gene Editing to Protein Optimization, which seems tailor-made for my interest in protein research. At the same time, I look forward to strengthening my knowledge of combinatorial algorithms, statistical theory, and probabilistic graphical models, which are all essential in building cutting edge AI products given the probabilistic nature and complexity of biological data. I am keen to explore the possibility of conducting protein modeling, folding, and optimization research with Professor Jane Foster’s group and the SAIR Lab. Her research to identify drug candidates with proxy regression models instead of physics simulations (such as Marinus) deeply interests me. The opportunity to work with Professor Foster, other CS faculty, and my peers will be a valuable asset in my academic development.
We are only beginning to understand the impact of AI on healthcare. The inevitable integration of these fields will require scientists with thorough expertise in both. My extensive work at the intersection of healthcare and ML has equipped me with the skills needed to help bridge the gap between them. I am confident that an education at Stark will allow me to grow as a data scientist, and help make healthcare more affordable and accessible, thus improving patient outcomes in the future.
Doesn’t This Essay Break the Rules?
Yes, it does! Aaron’s essay only half-follows the WriteIvy SOP template. I always advise structuring your SOPs like this:
- Introductory Frame Narrative
- Why This Program
- Why I’m Qualified
- Close Frame Narrative
Aaron, however, inverted sections 2 and 3. His “Why I’m Qualified” section came first. And I allowed it!
Well, it’s not often I’m comfortable letting students change the timeless structure of our template. It’s worked for thousands of applicants worldwide. It’s worked at Harvard, Stanford, Oxford, and nearly any other top university you can name. It works for STEM PhD applicants and MFA Artists. Yet, Aaron had a unique situation.
Because he was making a big leap from Bioengineering to Computer Science, Aaron needed to show right away that he was seriously qualified. If this essay followed the standard structure, the first few paragraphs would seem naïve.
“Wait, you’re a bioengineer?” the reader would say. “And you want to study CS? Maybe you’re a little underqualified, amigo.”
Thus, before describing his CS/Machine Learning/AI research proposal, Aaron needed to show the reader that 1) he was a serious researcher, and 2) he had a strong CS background despite being a Bioengineer.
Should you do this?
No, absolutely not. This is literally the only time I’ve advised a student to do this. It’s an extremely rare case.
Even so, Aaron’s example reveals the key to a fantastic machine learning SOP. We have to possess real purpose. Aaron wants to “help bridge the gap” between AI and healthcare. This essay explains exactly how and why, and that’s the lesson you should take away: your SOP needs to explain how you’re going to achieve big goals.
But I’m Not Nearly as Amazing as Him!
Don’t worry, I’m not either. Aaron was truly an amazing applicant.
But remember that one year prior, Aaron was rejected everywhere he applied. Yet, he didn’t sit around crying, worrying, and refreshing GradCafe. He found a job as a research assistant. He took night classes in NLP and algorithms. And, he worked his butt off to write a powerful, persuasive SOP the second time around.
If anything, Aaron’s essay shows us how insanely competitive the top Computer Science programs are. It isn’t easy getting into top schools. They want the best of the best. But you can’t be the best if you don’t have a powerful focus for your career, and revamping your SOP is a great way to level up and find that focus.
Conclusion for Your Machine Learning SOP
- Establish a big goal.
- Use the SOP to explain exactly how you’re going to achieve it.
Remember that “acquiring advanced CS knowledge” is not a goal. It’s a tool. How are you going to use that tool? Why does the world need people like you? This is the persuasive argument you need to make in your SOP.
I’m grateful to Aaron for allowing me to republish his work and brag about his success. I hope it inspires you to chase huge goals, and to sit down and think about how you’re going to achieve them. If you can explain that in your SOP, you’re going to be a huge success too.
Need help finding purpose in your Machine Learning SOP? Let me know! I’ve got some tricks that will help.
Which big goals are you chasing? How is grad school going to help you do so?