Cutting time, but not cutting corners: how AI is making disease treatment more effective
Written by Yumiko Imai ‘26
Edited by David Han ‘24
With the rise of ChatGPT, the effects of artificial intelligence on academia have come to the forefront of mainstream media regarding the authenticity of work. It has even further prompted questions about the increasing role of AI in our society as a whole and how we will adapt to similar new technologies that are here to stay.
However, many academics, doctors, and pathologists are seeing the benefits of artificial intelligence. They are finding ways to use machine and deep learning to their advantage to speed up processes, supplement their work, and assist with treatment decision-making.
One promising field being revolutionized by AI is protein design and development. David Baker’s Lab at the University of Washington has used software that they developed over the past 30 years called Rosetta and RoseTTAFold to predict amino acid sequences, structure, and folding patterns (1). But, their use of these two alone were unable to synthesize a functional protein. So, they used machine-learning scientist Justas Dauparas’ network called ProteinMPNN which can tweak the amino acid sequences to make the proteins functional (1). This highlights the collaborative nature of this field. These groups of scientists spent years developing their own AI tools and then shared them with others who have similar research topics. They have a shared view of the possibilities of AI and a common goal of scientific advancement.
The proteins created through the use of AI often have complex symmetry and interesting structures which are very different from proteins that can be found in nature. These proteins are synthesized with certain tasks in mind. AI will create a specific sequence which can then be embedded into a previously synthesized protein (1). This allows enzymes to be designed that catalyze a particular reaction or proteins to be made that have the potential to be used in a vaccine against a certain virus. Ali Madani, former research scientist at Salesforce Research and a primary author of the paper “Large Language Models Generate Functional Protein Sequences Across Diverse Families,” in collaboration with University of California - San Francisco, said that “the capability to generate functional proteins from scratch out-of-the-box demonstrates we are entering into a new era of protein design. This is a versatile new tool available to protein engineers, and we’re looking forward to seeing the therapeutic applications'' (2).
AI protein development is paving a path for itself and proving that it will play an important therapeutic role in the future. One promising drug called SGR-1505 was developed to treat B-cell lymphoma. It was developed by Schrodinger, Inc., which is a scientific software company. Their machine-learning platform sorted through over 8 billion compounds over 10 months to ultimately identify SGR-1505, a protein that inhibits MALT1 (3). MALT1 is a protein that activates immune cells and its abnormal expression is closely associated with autoimmune diseases and lymphomas, including B-cell lymphoma (4). SGR-1505’s preclinical data was promising and it was approved in late 2022 by the FDA to initiate a Phase 1 clinical trial (5). This is just one of many drugs developed by AI that are currently in various stages of clinical trials, but it gives insight into the current state and future possibilities of AI protein design.
Pivoting from drug development to treatment delivery and care, AI has the potential to help providers with decision-making. Deep learning models have started to be used in oncology, cardiology, and pathology. AI is proving itself to be very good at predicting patient outcomes and identifying optimal treatment routes due to its ability to discover patterns in patients and the course of disease (6).
At the University of Waterloo, researchers have developed Cancer-Net SCa, a deep learning network that is focused on skin cancer detection. It is the first AI created that can detect skin cancer from dermoscopy images. The researchers’ analysis of Cancer-Net SCa’s performance included that it identified “diagnostically relevant critical factors” and leveraged them over “irrelevant visual indicators” (7). The significance of this creation is that skin cancer is the most frequently diagnosed form of cancer in the United States. Patient outcomes are highly dependent on early detection and effective screening, so an AI that can assist dermatologists in ensuring that observation is all-encompassing and thorough has positive implications for patients (7).
More recently, a group of researchers from the same institution have developed deep-learning AI that can predict if women with breast cancer would benefit from chemotherapy prior to treatment (6). Neoadjuvant chemotherapy is a treatment that is done prior to surgery that can reduce invasiveness and make surgeries easier. For some patients, this proves to be effective in the long run (8). In other cases, it is not a suitable treatment path. This AI uses MRI, other imaging, and outcomes information from past breast cancer cases to make decisions about the course of treatment that would be most conducive to their situation (6).
At the center of both projects at the University of Waterloo is the mission to improve the patient treatment experience. It prioritizes efficiency and effective surgery and chemotherapy practices. The special part of these developments is that they are open-source (7). As an ongoing initiative to increase collaboration between researchers and clinicians, these AI models are available for public use.
Revolutionary discoveries are being made in the field of cancer research regarding protein targets and tailored treatment. In the clinical setting where patients are interacting with doctors and receiving care, there are new opportunities for clinicians to be sure that they are making the best decisions for their patients. On both fronts, AI is making its mark. It is remarkable that this technology is beginning to play a larger and larger role in medical treatment today.
1. Callaway E. Scientists are using AI to dream up revolutionary new proteins. Nature. 2022 Sep 15;609(7928):661–2.
2. Madani A, Krause B, Greene ER, Subramanian S, Mohr BP, Holton JM, et al. Large language models generate functional protein sequences across diverse families. Nat Biotechnol. 2023 Jan 26;1–8.
3. Hale C. Schrödinger to kick off first human trial of its computer-designed blood cancer drug [Internet]. Fierce Biotech. 2022 [cited 2023 Mar 6]. Available from: https://www.fiercebiotech.com/medtech/schrodinger-launch-first-human-trial-its-computer-designed-blood-cancer-drug
4. Liang X, Cao Y, Li C, Yu H, Yang C, Liu H. MALT1 as a promising target to treat lymphoma and other diseases related to MALT1 anomalies. Med Res Rev. 2021 Jul;41(4):2388–422.
5. Schrödinger Announces FDA Clearance of Investigational New Drug Application for SGR-1505, a MALT1 Inhibitor [Internet]. 2022 [cited 2023 Mar 6]. Available from: https://www.businesswire.com/news/home/20220628005165/en/Schr%C3%B6dinger-Announces-FDA-Clearance-of-Investigational-New-Drug-Application-for-SGR-1505-a-MALT1-Inhibitor
6. Tai C en A, Hodzic N, Flanagan N, Gunraj H, Wong A. Cancer-Net BCa: Breast Cancer Pathologic Complete Response Prediction using Volumetric Deep Radiomic Features from Synthetic Correlated Diffusion Imaging [Internet]. arXiv; 2022 [cited 2023 Feb 28]. Available from: http://arxiv.org/abs/2211.05308
7. Lee JRH, Pavlova M, Famouri M, Wong A. Cancer-Net SCa: tailored deep neural network designs for detection of skin cancer from dermoscopy images. BMC Med Imaging. 2022 Aug 9;22(1):143.
8. Neoadjuvant Chemotherapy | Breast Cancer Care | Mercy Health [Internet]. [cited 2023 Mar 6]. Available from: https://www.mercy.com/health-care-services/cancer-care-oncology/specialties/breast-cancer-treatment/treatments/neoadjuvant-chemotherapy
9. [Image] Proteins.jpeg [Internet] [cited 2023 March 6] Available from: https://www.nature.com/articles/d41586-022-02947-7