Abstract
Fibrosis is a pathological process contributing to approximately 45 percent of mortality in the United States through various dysfunctions and fibrotic disorders, including cystic fibrosis, chronic pancreatitis, and viral hepatitis. In response to inflammation and injury, stimuli such as TGF-β trigger the differentiation of fibroblasts into activated myofibroblasts. In fibrosis, however, the fibroblast-to-myofibroblast transition (FMT) is dysregulated, leading to excessive extracellular matrix (ECM) deposition. The chronic over-proliferation of collagen from this process results in the thickening, scarring, and overhardening of tissue, reducing organ function over time.
Despite decades of research developing anti-fibrotic agents, very few drugs have been effective and approved by the FDA. The limited progress in treating fibrosis can be attributed to flaws in preclinical studies, particularly inconsistent and unstandardized methods in activated myofibroblast identification. Identification of activated fibroblasts remains largely manual and subjective, typically classifying myofibroblasts based on alpha-smooth muscle actin (α-SMA) expression. Furthermore, FMT identification has occurred mainly through a binary classification style, with options being either “activated” or “quiescent,” which fails to capture the complexities of myofibroblast activation.
Therefore, for our technical project, my partner and I worked to develop an automated image analysis pipeline to quantify fibroblast activation. Upon the development of this pipeline, we sought to measure FMT progression through immunofluorescence intensity and mathematical morphological descriptors, as well as identify filamentous actin morphological features that predict activation state.
The current failures and flaws of fibrosis and fibroblast research inspired the topic for the sociotechnical paper. I was particularly interested in how the activation state of the fibroblast was manual and subjective. Microscopy is used widely across scientific disciplines, which made me wonder how many other studies and projects had subjective interpretations of images. This led me to think deeper about how much personal input, opinion, and thought come into play in scientific research, which is often tightly controlled.
My STS paper looks into where subjectivity appears in experimental practices. I employ the sociology of scientific knowledge (SSK) framework to review current sentiments surrounding objectivity and identify where subjectivity is present, even when it is believed that it isn’t there. To do this, I draw from Theodore Porter’s Trust in Numbers: The Pursuit of Objectivity in Science and Public Life, where he explains how the different types of objectivity are ineffective methods of removing human judgment from the scientific process. I also use Michael Mulkay’s critique of scientific Mertonian norms, where he describes how Robert Merton’s informal code of science has more nuance than believed, and it is not truly followed.
I then identify instances that are commonly perceived to be objective and apply the previously stated analytical frameworks to unveil where the subjectivity resides. These instances can be categorized into three main groups: mentorship and tacit knowledge, measurements and numbers, and peer review and institutions. Under each of the sections, I address how algorithms and automated systems are believed to solve the hidden presence of subjectivity, and how, in each scenario, AI and automated systems fail to provide the objectivity that many of the population believe it does.
Although the paper critiques the limitations of AI as a solution to subjectivity, it also considers pathways forward. In particular, I emphasize that the goal should not be to eliminate subjectivity, but to acknowledge and manage it. Recognizing subjectivity as an inherent feature of scientific practice allows for the development of strategies that promote transparency, accountability, and more responsible interpretation of results, both in traditional experimental methods and in the design of automated systems.
