Megan Mayerle, Ph.D. and Helen M. Blau, Ph.D. (PAS)

Abstract

In recent years, the application of Artificial Intelligence (AI) has profoundly impacted biomedical research (Topol, 2019). AI algorithms, particularly machine learning and deep learning models, can analyze large volumes of complex biological data more efficiently than traditional methods. AI also facilitates the integration of various omics data (epigenomics, transcriptomics, proteomics) to provide a comprehensive understanding of complex biological processes. By combining data from different sources, AI models can uncover how molecular pathways interact and identify patterns and correlations that may not be evident through manual analysis, leading to new insights. AI has also greatly improved microscopy-based studies and other imaging techniques. AI is proving invaluable to medicinal chemists in the design of therapeutics, speeding drug discovery and reducing costly testing of a multiplicity of related compounds in animals.

Here we provide specific examples of AI applications to biomedical research from our laboratory. The first three were instrumental to our studies of a gerozyme, an enzyme we discovered that increases with aging and is a pivotal molecular determinant of the devastating muscle wasting that leads to sarcopenia, and the fourth is an application of utility to physicians in underdeveloped countries for the diagnosis of melanoma. (1) We developed AI-powered image analysis of timelapse videos of single cells in order to track cell movement, divisions, fusions, death and gene expression that allowed us to generate information regarding cell fate decisions and lineage trees. Such algorithms can aid evaluation of drugs for their potential to increase stem cell viability, stemness properties and cell divisions, and their potential to augment regeneration after injury or exercise. (2) AI algorithms for image analysis are revolutionizing the speed and accuracy of previously labor-intensive evaluation of tissue section histology essential to laboratory research and medical diagnoses. For example, in lieu of the traditional 3 markers per tissue section, currently as many as 60 markers can be analyzed simultaneously using AI algorithms to resolve a multiplicity of cell types, their relationship to one another in cellular neighborhoods, and gene expression patterns, providing unprecedented insights into development and pathogenesis, and yielding more accurate diagnoses. (3) Understanding the mechanisms that control gene expression is not only of fundamental interest, but key to targeting genes with therapeutics based on CRISPR technologies, which can render the genes active or inactive, as desired. AI tools are able to uncover DNA motifs and predict protein transcription factor interactions at these motifs, which is not possible using traditional experimental approaches. (4) Beyond the lab, AI-assisted medical imaging analysis has emerged as a potentially lifesaving tool for clinicians (Topol, 2019). We provide an example of AI-assisted diagnosis of melanoma we developed together with dermatologists and AI expert Professor Sebastian Thrun at Stanford to aid physicians in rural areas and in underdeveloped countries in assessing if a skin lesion is potentially lethal and warrants further medical attention. These examples provide a glimpse of the vast potential of AI to accelerate the pace of research, offer new tools and approaches for understanding basic biological mechanisms, their application to disease, and the development of diagnostics and therapeutics, and their translation to the clinic.

Introduction

Sarcopenia is the clinically defined loss of muscle mass and strength with aging that poses major health risks and compromises quality of life(Cruz-Jentoft et al., 2019). Sarcopenic individuals have increased susceptibility to falls and fractures, and difficulty recovering from injuries. This can lead to a downward spiral of increasing pain and reduced physical activity, that further exacerbates muscle loss and functional decline, as everyday tasks become more challenging and physical limitations increase, culminating in a loss of independence. Sarcopenia is often associated with metabolic disorders, cardiovascular complications, and respiratory problems which combine to increase overall risk of mortality(Larsson et al., 2019). Our lab recently discovered a new hallmark of aging, which we termed a “gerozyme”, a new class of enzyme that plays a critical role in the aging process and the maintenance of cellular function (Fig 1) (Bakooshli et al., 2023; Ho et al., 2017; Palla et al., 2021). Gerozymes impact various biochemical pathways that regulate tissue repair and overall metabolic health, and potentially longevity. Gerozymes contribute to oxidative stress, chronic inflammation, and cellular senescence, which are key factors in aging and age-related diseases. Our lab identified 15-hydroxyprostaglandin dehydrogenase (15-PGDH) as a muscle gerozyme (Bakooshli et al., 2023; Palla et al., 2021). 15-PGDH is a catabolic enzyme that increases in expression with age and degrades prostaglandins including prostaglandin E2 (PGE2), a lipid metabolite. The potency of the gerozyme to impact aging derives from its multiple targets: muscle stem cells, muscle fibers and neuromuscular connectivity (Fig 2).

Muscle stem cells. Muscle stem cells are critical to tissue regeneration after damage, which can be caused by exercise and is the means by which exercise leads to increased muscle mass and force. We discovered that PGE2 is required for muscle regeneration as it is essential to stem cell viability and proliferative capacity (Bakooshli et al., 2023; Cheng et al., 2021; Palla et al., 2021). Experiments performed in mice show that stem cells that lack the receptor that mediates PGE2 signaling cannot meet the needs for muscle repair and after an injury, and mice become weaker. Similarly, if after an injury, mice are treated with NSAIDs (non-steroidal anti-inflammatory agents like ibuprofen) which block endogenous PGE2 synthesis, mice similarly lose strength. These data were depicted in the New York Times when our paper was published in 2017 with the caption “no pain, no gain” – ingestion of an NSAID will negate the benefit to the muscles of running a marathon or working out in a gym.

With aging, muscle stem cell function declines, leading to a marked decline in the muscle's ability to repair and regenerate following injury. This aging-associated loss of regenerative capacity has been attributed to multiple factors including the breakdown of quiescence, mitotic catastrophe, aberrant transition to commitment, and chromatin modifications (Blau et al., 2015; Brunet et al., 2022; Fuchs & Blau, 2020; Hwang & Brack, 2018; Porpiglia & Blau, 2022). The etiology notwithstanding, the outcome is clear: with aging there is a profound reduction in the number and function of muscle stem cells (Blau et al., 2015; Cosgrove et al., 2014; Sousa-Victor et al., 2022). Additionally, with aging the muscle microenvironment becomes less conducive to regeneration due to increased fibrosis, chronic low-grade inflammation, and altered extracellular matrix composition(Fuchs & Blau, 2020). These systemic changes also affect intercellular signaling pathways, further compromising the cellular composition and thus regenerative potential of the muscle(Porpiglia & Blau, 2022). As a result, the aged muscle is less efficient in recovering from injuries, which contributes to the progressive decline in muscle mass and function commonly observed in elderly individuals.

Myofiber integrity. PGE2 levels are also crucial to myofiber function. With aging PGE2 declines due to an increase in the expression of the gerozyme that degrades it. The reduction in PGE2 has profound effects on muscle tissue structure and function(Palla et al., 2021). We found that reducing PGE2 levels in young mice by overexpressing 15-PGDH impaired muscle function, leading to muscle atrophy. Within one month, the muscles of young mice shrank and weakened. Conversely, inhibiting 15-PGDH activity in aged mice significantly increased muscle mass, strength and endurance running on a treadmill. At the molecular level, the mitochondria, the organelles responsible for energy production, which become increasingly sparse and dysfunctional with aging leading to lower energy availability and increased oxidative stress, were increased in number and function following 15-PGDH inhibition. In addition, deleterious pathways were suppressed, for example the aging associated increase in protein degradation by the products of atrogenes, ubiquitin ligases, and TGF-β signaling pathway that increases fibrosis and tissue stiffness (Palla et al., 2021; Wiedmer et al., 2021).

Innervation by motor neurons. Stimulation by neurons is critically important to muscle function. With aging, skeletal muscles lose neuromuscular connectivity, which accelerates muscle wasting and weakness. The fiber type composition of aged muscle changes with more prevalent type 1 small, slow contractile, low tension, highly aerobic myofibers and comparatively fewer type 2 larger, fast contractile, high tension, quick fatigue myofibers (Siparsky et al., 2014). This change accounts for the progressive inability of aged individuals to perform fast high-powered movements like sprinting. Two interventions have been shown to be efficacious in preventing or reversing the loss of innervation with aging in mice. A life-long calorie-restricted diet decreased the incidence of pre- and post-synaptic abnormalities in old age. One month of exercise (wheel running) of aged mice reduced the synaptic changes that had accumulated (Valdez et al., 2010). However, such interventions can be difficult for aged humans to implement. Drugs that can restore innervation are lacking. We found that 15-PGDH inhibition significantly restored innervation to aged myofibers serving as an exercise mimetic, providing an additional mechanism accounting for the increase in strength we previously reported (Bakooshli et al., 2023).

Together, these data provide insights into the mechanism by which PGE2 and inhibition of the gerozyme impacts not only muscle stem cell behavior, but also the muscle myofiber and the connection between muscles and motor neurons. Modulating 15-PGDH activity therefore represents an innovative approach to enhance muscle repair, and a novel therapeutic strategy to treat age-related muscle wasting and sarcopenia (Bakooshli et al., 2023; Ho et al., 2017; Palla et al., 2021). Below we highlight how AI has contributed to these discoveries.

1. AI resolves how PGE2 impacts muscle stem behavior. In muscle regeneration, muscle stem cells must self-renew in order to maintain the stem cell pool and commit to creating specialized progeny to regenerate and build the tissue. This entails both asymmetric and symmetric cell divisions (Chargé & Rudnicki, 2004). Of great interest are regulators that can enhance stem cell numbers and function.

To establish the mechanism by which growth factors or metabolites impact stem cell fate, single cell timelapse analysis is a robust tool. However, the analysis of the images captured in hundreds of videos is a highly labor intensive and slow process. We therefore sought an AI algorithm to simplify and accelerate the analysis of these movies. Notably, training a computer to recognize cell division and cell death events in moving cells in real time based on morphology is not trivial. This type of single cell analysis is key, as population-level analyses can mask contributions of cell subpopulations and dynamics of cell fate changes (Schroeder, 2011). Working with world renowned Stanford AI expert Sebastian Thrun, who played a major role in the development of Google’s driverless cars, we identified a talented Swedish student, Klas Magnusson, capable of meeting this challenge. Using his algorithm, which has now won international prizes for 13 stem cell types (Gilbert et al., 2010; Togninalli et al., 2023; Ulman et al., 2017) we were able to assess the impact of PGE2 and other growth factors on the fate of single muscle stem in real time. Using markers that could distinguish stem from committed progeny, our data showed that PGE2 significantly increases stem cell numbers by increasing cell division and viability and shifts cells toward a more stem cell-like phenotype compared to controls (Ho et al., 2017; Togninalli et al., 2023). By day 3, muscle stem cell numbers had tripled, and by day 7, the proportion of stem cells present was markedly higher in the PGE2-treated population, showing that PGE2 treatment enhances the retention of a stem cell phenotype while increasing stem cell numbers (Togninalli et al., 2023) (Fig 3).

We are located in the Baxter Laboratory and hence designated this AI method as the Baxter Algorithm (Magnusson et al., 2015). The Baxter Algorithm tracks individual cells through successive images of a timelapse video. Using the Baxter Algorithm, we categorized individual cells at each point as 1) not dividing, 2) dividing to make one stem cell and one differentiated cell, 3) dividing to make two stem cells, 4) switching from a stem cell to a differentiated cell, or 5) dying. With these classifications, we were able to construct complete lineage trees for each muscle stem cell providing deeper insights into stem cell survival and fate.

We applied the Baxter Algorithm to understand how PGE2 impacts muscle stem cell behavior. Our optimized AI analysis showed that PGE2-treated stem cells underwent more division events compared to controls and were more likely to divide to create two new stem cells (symmetric self-renewal). PGE2-treated cells divided more often and lived longer than untreated MuSCs (Togninalli et al., 2023). Thus, our novel AI approach showed that PGE2 expands the stem cell pool by increasing self-renewing divisions and promoting stem cell survival.

However, muscle stem cells are also exquisitely sensitive to their physical environment. During regeneration, muscle stem cells receive cues from their microenvironment that guide their expansion, differentiation, and return to a quiescent state. Understanding the complex interplay between these biophysical and biochemical signals is crucial for developing therapies that target or utilize stem cells. To investigate the role of mechanical cues provided by the extracellular matrix, we use chemically defined hydrogel substrates with adjustable stiffness and adhesive ligand composition to study how muscle stem cells respond to matrix signals during the early and late stages of regeneration. We applied the Baxter Algorithm to track muscle stem cells on these substrates and chronicle their response. Our findings show that softer hydrogels, mimicking the stiffness of healthy muscle, promote stem cell expansion and differentiation, while stiffer hydrogels hinder these fate decisions. We also discovered that factors like PGE2 work synergistically with physical cues to enhance stem cell expansion on soft substrates and inhibit myogenic progression on stiff substrates (Madl et al., 2021). To assess whether changes in matrix stiffness over time, similar to those in the regenerating microenvironment, affect stem cell fate, we developed a photo-responsive hydrogel system that can be softened or stiffened on demand. Muscle stem cells cultured on these materials revealed that the cellular response to a stiff microenvironment is determined within the first three days of culture and that stem cells harbor a mechanical memory for substrate stiffness (Madl et al., 2021). These results, only possible because of recent advances in AI-facilitated cell tracking and classification, underscore the significance of temporally controlled biophysical and biochemical cues in regulating muscle stem cell fate, which can be leveraged to enhance regenerative medicine strategies for restoring skeletal muscle tissue and extended to other tissue stem cell types.

2. AI augments analysis of tissue sections to chart changes in cellular neighborhoods

With aging, muscle mass and function diminish, and the cellular composition of muscle tissue becomes increasingly heterogeneous (Larsson et al., 2019). How cells interact within a tissue directly affects how the tissue functions. To understand how increased levels of the gerozyme 15-PGDH could contribute to the decline in muscle function in aged and sarcopenic individuals, we needed a way to identify specific cell types that express 15-PGDH and lead to the decreased prostaglandin signaling that characterizes aged muscle tissue in relation to other cell types within the tissue. Traditional microscopy methods to identify cell types use specific fluorophores to label cells. There are a finite number of fluorophores available, and many of them exhibit significant background and spectral overlap, which limits the number of markers that can be tracked. To address this need and resolve how cells spatially interact with other cell types in the context of the skeletal muscle tissue microenvironment in aging, we optimized multiplex tissue imaging (CODEX, CO-Detection by indEXing)(Black et al., 2021) and applied it to the study of skeletal muscle. CODEX enables us to distinguish up to 60 proteins and 25 cell types on single sections of skeletal muscle via iterative imaging of the same tissue section (Fig 4). We use 3D single cell segmentation and high dimensional clustering using algorithms developed in our laboratory to detect the wide range of cell types found in muscle while retaining their spatial positions. CODEX enables unprecedented resolution of previously immeasurable cell-cell interactions and changes in tissue architecture.

The identification of cell types from imaging data is a computational challenge. Multiplex imaging approaches have lower dimensionality, imaging artifacts, and variable tissue autofluorescence which can contribute to poor performance in traditional clustering algorithms used to identify different cell types, requiring scientists to manually merge hundreds of clusters to obtain biologically relevant cell types. This is a major source of incorrect cell type annotation. To counter this, we developed HFCluster (Fig 5), an AI powered clustering pipeline optimized for multiplexed imaging data that overcomes the contribution of non-specific signal and noise in the clustering steps (Palla et al., 2021). HFCluster reduces the time needed to manually validate cell clusters from weeks to hours. Importantly, HFCluster identifies biologically relevant cell types reproducibly across experiments. By applying AI-powered HFCluster to our CODEX data, we identified myofibers and macrophages as the primary sites of the prostaglandin degradation that drives the dysfunction in aged muscles. This technology is enabling us to identify cellular neighborhoods and how they change with aging and disease and in response to drug treatments such as the small molecule that inhibits the gerozyme.

3. AI predicts protein interactions at regulatory motifs that drive gene expression

Aging is associated with changes to the epigenome. We hypothesized that PGE2 could exert its function by reversing specific changes to the muscle stem cell epigenome, impacting gene expression, e.g. of 15-PGDH, and therefore stem cell function. Chromatin accessibility refers to how accessible (open or closed) DNA is at specific sequences, or motifs. This affects gene expression by allowing or blocking access of transcription factors to genes. We isolated muscle stem cells from young and aged mice, treated them with PGE2 or a control, and then performed a sequencing method, known as assay for transposase-accessible chromatin with sequencing (ATAC-seq) that allows us to examine changes in chromatin accessibility. PGE2 induced a variety of complex changes in chromatin accessibility, however we could not integrate these diverse changes to create a coherent model. We turned to AI to help us understand this complex dataset.

We formed a collaboration with Anshul Kundaje, a Professor at Stanford. Anshul’s group created a deep learning model called ChromBPNet (Fig 6)(Avsec et al., 2021; Nair et al., 2023). ChromBPNet is a dual headed convolutional neural network model. It operates via two branches, with a locked branch trained on GC-matched non-regulatory regions of DNA to de-noise technical bias and a main branch that uses the ChromBPNet architecture to learn the cellular grammar for each locus across entire genome. This AI method allowed us to denoise our data by performing in silico mutagenesis and quantifying the importance of each base pair to the overall accessibility of local chromatin. ChromBPNet also learns transcription factor logic, the sequences and motif preferences of each transcription factor. ChromBPnet identified the transcription factors that are potential mediators of the gene expression changes that accompany aging.

We tested whether these AI-predicted transcription factors were indeed responsible for the rejuvenating effect of PGE2 exposure on aged muscle stem cells. By examining the transcriptomes of young and aged stem cells that had or had not been treated with PGE2, we validated the AI’s predictions. This is a key example of how AI can facilitate the integration and analysis of a dataset that is too complex for a scientist to evaluate manually. Through this collaborative effort, we now understand how the chromatin of a muscle stem cell responds to PGE2, which DNA motifs close or open, and can correlate this data to understand how these changes in chromatin accessibility impact the ability of various transcription factors to interact with the DNA to control gene expression. AI correctly identified which of the nearly 2000 transcription factors in the cell are involved in regulating the muscle stem cell response to PGE2.

4. AI facilitates classification of skin cancers

AI has tremendous clinical potential to empower patients and improve public health, as exemplified by our recent application of this technology to the diagnosis of skin cancer, which could significantly aid diagnosis in rural and underdeveloped countries. In 2017 my lab collaborated with Sebastian Thrun on an effort to use AI to improve screening for melanoma, a particularly lethal skin cancer (Esteva et al., 2017). Skin cancer is typically diagnosed through visual methods (Nataren et al., 2023). This process starts with a clinical screening and may be followed by dermoscopic analysis, biopsy, and histopathological examination. Diagnostic outcomes often rely on the dermatologist’s expertise, and diagnosing skin cancer visually is highly subjective. Therefore, developing an automated classification method for skin cancer that offers greater accuracy, affordability, and speed could have a major impact, particularly in communities with less access to state-of-the-art healthcare, and in areas with less well-trained physicians. However, automated classification of skin lesions from images presents a significant challenge due to the subtle variations in their appearance.

Recent advancements in computation and access to large datasets have enabled deep learning algorithms to meet and sometimes surpass human performance in various visual tasks, including playing video games (Mnih et al., 2015), strategic board games like Go (Silver et al., 2016), and object recognition (Russakovsky et al., 2015). However, due to the complexity and variability of skin disease images, automating skin cancer classification is especially challenging, as skin lesions, even those of the same class, can differ significantly in color, features, structure, size, and location. Our goal was to develop a Convolutional Neural Network (CNN) that achieves performance on par with dermatologists in three critical diagnostic tasks: melanoma classification, melanoma classification using dermoscopy, and carcinoma classification based on image-based classification. We employed a GoogleNet Inception v3 CNN architecture and fine-tuned this model on our dataset using transfer learning such that the CNN trained on dermatologist-labeled images sourced from 18 clinician-curated, open-access online repositories, as well as clinical data from Stanford University Medical Center. We then created an algorithm to divide diseases into fine-grained training classes (such as amelanotic melanoma and acrolentiginous melanoma). During inference, the CNN generates a probability distribution over these fine classes. After validating this approach computationally, we tested its performance against the performance of 21 clinical dermatologists. Dermatologists were asked, based on image inspection, whether to biopsy/treat each lesion or reassure the patient. The CNN outperformed most dermatologists (Esteva et al., 2017).

This study demonstrates the effectiveness of AI and deep learning in dermatology, applying it to both general skin conditions and specific cancers. Our single convolutional neural network matched the performance of dermatologists. It is a fast, scalable method that can be deployed on a smartphone and has the potential for significant clinical impact, such as expanding primary care capabilities and enhancing decision-making for dermatology specialists. The ability to classify skin lesion images with the accuracy of a board-certified dermatologist could greatly increase access to essential medical care. Regrettably dissemination of this potent AI technology has been hindered, as investors do not see a viable commercial path forward.

Conclusion

Here we describe how our lab has employed AI to spur advances in both the laboratory and the clinic. We are just one group at one university. By harnessing the power of machine learning, deep learning, and other AI technologies, scientists and healthcare professionals can analyze vast datasets, accelerate drug discovery, improve diagnostic accuracy, and personalize medicine. The integration of AI into biomedical research and medicine is poised to revolutionize both the understanding of complex biological systems and the delivery of patient care in developed and underdeveloped countries.

Acknowledgements

We would like to thank current and previous members of the Blau lab, especially NAMES who made key contributions to this work. We also thank Dr. Garry Nolan and his group for assistance with CODEX, and Dr. Anshul Kundaje and his group for their work on ChromBPNet. Funding for this work was provided by the Baxter Foundation and the Li Ka Shing Foundation.

References

Avsec, Ž., Weilert, M., Shrikumar, A., Krueger, S., Alexandari, A., Dalal, K., Fropf, R., McAnany, C., Gagneur, J., Kundaje, A., & Zeitlinger, J. (2021). Base-resolution models of transcription-factor binding reveal soft motif syntax. Nature Genetics, 53(3), 354–366. https://doi.org/10.1038/s41588-021-00782-6

Bakooshli, M.A., Wang, Y.X., Monti, E., Su, S., Kraft, P., Nalbandian, M., Alexandrova, L., Wheeler, J.R., Vogel, H., & Blau, H.M. (2023). Regeneration of neuromuscular synapses after acute and chronic denervation by inhibiting the gerozyme 15-prostaglandin dehydrogenase. Science Translational Medicine, 15(717), eadg1485. https://doi.org/10.1126/scitranslmed.adg1485

Black, S., Phillips, D., Hickey, J.W., Kennedy-Darling, J., Venkataraaman, V.G., Samusik, N., Goltsev, Y., Schürch, C.M., & Nolan, G.P. (2021). CODEX multiplexed tissue imaging with DNA-conjugated antibodies. Nature Protocols, 16(8), 3802–3835. https://doi.org/10.1038/s41596-021-00556-8

Blau, H.M., Cosgrove, B.D., & Ho, A.T.V. (2015). The central role of muscle stem cells in regenerative failure with aging. Nature Medicine, 21(8), 854–862. https://doi.org/10.1038/nm.3918

Brunet, A., Goodell, M.A., & Rando, T.A. (2022). Ageing and rejuvenation of tissue stem cells and their niches. Nature Reviews. Molecular Cell Biology. https://doi.org/10.1038/s41580-022-00510-w

Chargé, S.B.P., & Rudnicki, M.A. (2004). Cellular and molecular regulation of muscle regeneration. Physiological Reviews, 84(1), 209–238. https://doi.org/10.1152/physrev.00019.2003

Cheng, H., Huang, H., Guo, Z., Chang, Y., & Li, Z. (2021). Role of prostaglandin E2 in tissue repair and regeneration. Theranostics, 11(18), 8836–8854. https://doi.org/10.7150/thno.63396

Cosgrove, B.D., Gilbert, P.M., Porpiglia, E., Mourkioti, F., Lee, S.P., Corbel, S.Y., Llewellyn, M.E., Delp, S.L., & Blau, H.M. (2014). Rejuvenation of the muscle stem cell population restores strength to injured aged muscles. Nature Medicine, 20(3), 255–264. https://doi.org/10.1038/nm.3464

Cruz-Jentoft, A.J., Bahat, G., Bauer, J., Boirie, Y., Bruyère, O., Cederholm, T., Cooper, C., Landi, F., Rolland, Y., Sayer, A.A., Schneider, S.M., Sieber, C.C., Topinkova, E., Vandewoude, M., Visser, M., Zamboni, M., & Writing Group for the European Working Group on Sarcopenia in Older People 2 (EWGSOP2), and the Extended Group for EWGSOP2. (2019). Sarcopenia: Revised European consensus on definition and diagnosis. Age and Ageing, 48(1), 16–31. https://doi.org/10.1093/ageing/afy169

Esteva, A., Kuprel, B., Novoa, R.A., Ko, J., Swetter, S.M., Blau, H.M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115–118. https://doi.org/10.1038/nature21056

Fuchs, E., & Blau, H.M. (2020). Tissue Stem Cells: Architects of Their Niches. Cell Stem Cell, 27(4), 532–556. https://doi.org/10.1016/j.stem.2020.09.011

Gilbert, P., Havenstrite, K., Magnusson, K., Sacco, A., Leonardi, N., Kraft, P., Nguyen, N., Thrun, S., Lutolf, M., & Blau, H. (2010). Substrate elasticity regulates skeletal muscle stem cell self-renewal in culture. Science (New York, N.Y.), 329(5995), 1078–1081. https://doi.org/10.1126/science.1191035

Ho, A.T.V., Palla, A.R., Blake, M.R., Yucel, N.D., Wang, Y.X., Magnusson, K.E.G., Holbrook, C.A., Kraft, P.E., Delp, S.L., & Blau, H.M. (2017). Prostaglandin E2 is essential for efficacious skeletal muscle stem-cell function, augmenting regeneration and strength. Proceedings of the National Academy of Sciences of the United States of America, 114(26), 6675–6684. https://doi.org/10.1073/pnas.1705420114

Hwang, A.B., & Brack, A.S. (2018). Muscle Stem Cells and Aging. Current Topics in Developmental Biology, 126, 299–322. https://doi.org/10.1016/bs.ctdb.2017.08.008

Larsson, L., Degens, H., Li, M., Salviati, L., Lee, Y.I., Thompson, W., Kirkland, J.L., & Sandri, M. (2019). Sarcopenia: Aging-Related Loss of Muscle Mass and Function. Physiol Rev, 99(1), 427–511. https://doi.org/10.1152/physrev.00061.2017

Madl, C.M., Flaig, I.A., Holbrook, C.A., Wang, Y.X., & Blau, H.M. (2021). Biophysical matrix cues from the regenerating niche direct muscle stem cell fate in engineered microenvironments. Biomaterials, 275, 120973. https://doi.org/10.1016/j.biomaterials.2021.120973

Magnusson, K.E.G., Jalden, J., Gilbert, P.M., & Blau, H.M. (2015). Global linking of cell tracks using the Viterbi algorithm. IEEE Transactions on Medical Imaging, 34(4), 911–929. https://doi.org/10.1109/TMI.2014.2370951

Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A.A., Veness, J., Bellemare, M.G., Graves, A., Riedmiller, M., Fidjeland, A.K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A., Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg, S., & Hassabis, D. (2015). Human-level control through deep reinforcement learning. Nature, 518(7540), 529–533. https://doi.org/10.1038/nature14236

Nair, S., Ameen, M., Sundaram, L., Pampari, A., Schreiber, J., Balsubramani, A., Wang, Y.X., Burns, D., Blau, H.M., Karakikes, I., Wang, K.C., & Kundaje, A. (2023). Transcription factor stoichiometry, motif affinity and syntax regulate single-cell chromatin dynamics during fibroblast reprogramming to pluripotency. bioRxiv: The Preprint Server for Biology, 2023.10.04.560808. https://doi.org/10.1101/2023.10.04.560808

Nataren, N., Yamada, M., & Prow, T. (2023). Molecular Skin Cancer Diagnosis: Promise and Limitations. The Journal of Molecular Diagnostics: JMD, 25(1), 17–35. https://doi.org/10.1016/j.jmoldx.2022.09.008

Palla, A.R., Ravichandran, M., Wang, Y.X., Alexandrova, L., Yang, A.V., Kraft, P., Holbrook, C.A., Schürch, C.M., Ho, A.T.V., & Blau, H.M. (2021). Inhibition of prostaglandin-degrading enzyme 15-PGDH rejuvenates aged muscle mass and strength. Science (New York, N.Y.), 371(6528), eabc8059. https://doi.org/10.1126/science.abc8059

Porpiglia, E., & Blau, H.M. (2022). Plasticity of muscle stem cells in homeostasis and aging. Current Opinion in Genetics & Development, 77, 101999. https://doi.org/10.1016/j.gde.2022.101999

Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., Berg, A.C., & Fei-Fei, L. (2015). ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision, 115(3), 211–252. https://doi.org/10.1007/s11263-015-0816-y

Schroeder, T. (2011). Long-term single-cell imaging of mammalian stem cells. Nature Methods, 8(4 Suppl), S30-35. https://doi.org/10.1038/nmeth.1577

Silver, D., Huang, A., Maddison, C.J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., Dieleman, S., Grewe, D., Nham, J., Kalchbrenner, N., Sutskever, I., Lillicrap, T., Leach, M., Kavukcuoglu, K., Graepel, T., & Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484–489. https://doi.org/10.1038/nature16961

Siparsky, P.N., Kirkendall, D.T., & Garrett, W.E. (2014). Muscle Changes in Aging. Sports Health, 6(1), 36–40. https://doi.org/10.1177/1941738113502296

Sousa-Victor, P., García-Prat, L., & Muñoz-Cánoves, P. (2022). Control of satellite cell function in muscle regeneration and its disruption in ageing. Nature Reviews Molecular Cell Biology, 23(3), Article 3. https://doi.org/10.1038/s41580-021-00421-2

Togninalli, M., Ho, A.T.V., Madl, C.M., Holbrook, C.A., Wang, Y.X., Magnusson, K.E.G., Kirillova, A., Chang, A., & Blau, H.M. (2023). Machine learning-based classification of dual fluorescence signals reveals muscle stem cell fate transitions in response to regenerative niche factors. NPJ Regenerative Medicine, 8(1), 4. https://doi.org/10.1038/s41536-023-00277-4

Topol, E.J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44–56. https://doi.org/10.1038/s41591-018-0300-7

Ulman, V., Maška, M., Magnusson, K.E.G., Ronneberger, O., Haubold, C., Harder, N., Matula, P., Matula, P., Svoboda, D., Radojevic, M., Smal, I., Rohr, K., Jaldén, J., Blau, H.M., Dzyubachyk, O., Lelieveldt, B., Xiao, P., Li, Y., Cho, S.-Y., … Ortiz-de-Solorzano, C. (2017). An objective comparison of cell-tracking algorithms. Nature Methods, 14(12), 1141–1152. https://doi.org/10.1038/nmeth.4473

Valdez, G., Tapia, J.C., Kang, H., Clemenson, G.D., Gage, F.H., Lichtman, J.W., & Sanes, J.R. (2010). Attenuation of age-related changes in mouse neuromuscular synapses by caloric restriction and exercise. Proceedings of the National Academy of Sciences, 107(33), 14863–14868. https://doi.org/10.1073/pnas.1002220107

Wiedmer, P., Jung, T., Castro, J.P., Pomatto, L.C.D., Sun, P.Y., Davies, K.J.A., & Grune, T. (2021). Sarcopenia – Molecular mechanisms and open questions. Ageing Research Reviews, 65, 101200. https://doi.org/10.1016/j.arr.2020.101200