Job Title
Research Fellow
Academic Rank
Fellow or Postdoc
Department
Pathology
Authors
Muhammad Shaban, Wiem Lassoued, Kenneth Canubas, Shania Bailey, Clint Allen, Julius Strauss, James L Gulley, Sizun Jiang, Faisal Mahmood, George Zaki, and Houssein A Sater
Principal Investigator
Faisal Mahmood
Categories
Tags
Multiplexed imaging has revolutionized the visualization of numerous protein markers within tissue samples. Nevertheless, the elevated marker count amplifies the risk of staining mishaps, resulting in escalated resource consumption during multiplexed image acquisition, primarily due to the need for restaining. To tackle this challenge, we propose an innovative approach: MAXIM (MArker imputation model for multipleXed IMages), harnessing the power of deep learning. MAXIM adeptly predicts missing protein markers by exploiting latent biological connections between them. Our model undergoes comprehensive evaluation at both pixel and cell levels, spanning various cancer types. Notably, MAXIM exhibits exceptional performance in cell classification. One of its key merits lies in bolstering interpretability. This is achieved by gaining nuanced insights into the distinctive roles individual markers play in the imputation process. In practical implementation, MAXIM has the potential to substantially curtail the expenses and time associated with multiplexed image acquisition. By accurately predicting protein markers influenced by staining anomalies, MAXIM alleviates the necessity for repeated staining procedures. This innovative solution stands to transform the landscape of multiplexed imaging, making it more efficient and cost-effective.
Multiplexed imaging is a technique that lets us look at multiple proteins in a single tissue sample. But sometimes, when we try to see all these proteins simultaneously, some might not show up clearly. This means we need to redo the staining process, which takes up a lot of resources like time and materials. To solve this problem, we devised a computer program called MAXIM. This program uses deep learning, a bit like how our brains learn from experience, to figure out how these proteins are related to each other. It can look at the proteins that didn’t show up well and make a good guess about what they should look like based on the other proteins. We tested MAXIM on many images of tissues from different types of cancer. It’s good at guessing what the missing proteins should look like, even at the tiniest level. This helps us understand the cells better and even classify them accurately. MAXIM also tells us which proteins are most important for making these guesses. In real life, MAXIM could save time and money by helping us avoid redoing the staining process.