Chemistry-First Approach for Nomination of Personalized Treatment in Lung Cancer.

Diversity in the genetic lesions that cause cancer is extreme. In consequence, a pressing challenge is the development of drugs that target patient-specific disease mechanisms. To address this challenge, we employed a chemistry-first discovery paradigm for de novo identification of druggable targets linked to robust patient selection hypotheses. In particular, a 200,000 compound diversity-oriented chemical library was profiled across a heavily annotated test-bed of >100 cellular models representative of the diverse and characteristic somatic lesions for lung cancer. This approach led to the delineation of 171 chemical-genetic associations, shedding light on the targetability of mechanistic vulnerabilities corresponding to a range of oncogenotypes present in patient populations lacking effective therapy. Chemically addressable addictions to ciliogenesis in TTC21B mutants and GLUT8-dependent serine biosynthesis in KRAS/KEAP1 double mutants are prominent examples. These observations indicate a wealth of actionable opportunities within the complex molecular etiology of cancer.

DNA Methylation Regulates Alternative Polyadenylation via CTCF and the Cohesin Complex.

Dysregulation of DNA methylation and mRNA alternative cleavage and polyadenylation (APA) are both prevalent in cancer and have been studied as independent processes. We discovered a DNA methylation-regulated APA mechanism when we compared genome-wide DNA methylation and polyadenylation site usage between DNA methylation-competent and DNA methylation-deficient cells. Here, we show that removal of DNA methylation enables CTCF binding and recruitment of the cohesin complex, which, in turn, form chromatin loops that promote proximal polyadenylation site usage. In this DNA demethylated context, either deletion of the CTCF binding site or depletion of RAD21 cohesin complex protein can recover distal polyadenylation site usage. Using data from The Cancer Genome Atlas, we authenticated the relationship between DNA methylation and mRNA polyadenylation isoform expression in vivo. This DNA methylation-regulated APA mechanism demonstrates how aberrant DNA methylation impacts transcriptome diversity and highlights the potential sequelae of global DNA methylation inhibition as a cancer treatment.

Diagnosis of Alzheimer Disease and Tauopathies on Whole-Slide Histopathology Images Using a Weakly Supervised Deep Learning Algorithm.

Neuropathologic assessment during autopsy is the gold standard for diagnosing neurodegenerative disorders. Neurodegenerative conditions, such as Alzheimer disease (AD) neuropathological change, are a continuous process from normal aging rather than categorical; therefore, diagnosing neurodegenerative disorders is a complicated task. We aimed to develop a pipeline for diagnosing AD and other tauopathies, including corticobasal degeneration (CBD), globular glial tauopathy, Pick disease, and progressive supranuclear palsy. We used a weakly supervised deep learning-based approach called clustering-constrained-attention multiple-instance learning (CLAM) on the whole-slide images (WSIs) of patients with AD (n = 30), CBD (n = 20), globular glial tauopathy (n = 10), Pick disease (n = 20), and progressive supranuclear palsy (n = 20), as well as nontauopathy controls (n = 21). Three sections (A: motor cortex; B: cingulate gyrus and superior frontal gyrus; and C: corpus striatum) that had been immunostained for phosphorylated tau were scanned and converted to WSIs. We evaluated 3 models (classic multiple-instance learning, single-attention-branch CLAM, and multiattention-branch CLAM) using 5-fold cross-validation. Attention-based interpretation analysis was performed to identify the morphologic features contributing to the classification. Within highly attended regions, we also augmented gradient-weighted class activation mapping to the model to visualize cellular-level evidence of the model's decisions. The multiattention-branch CLAM model using section B achieved the highest area under the curve (0.970 ± 0.037) and diagnostic accuracy (0.873 ± 0.087). A heatmap showed the highest attention in the gray matter of the superior frontal gyrus in patients with AD and the white matter of the cingulate gyrus in patients with CBD. Gradient-weighted class activation mapping showed the highest attention in characteristic tau lesions for each disease (eg, numerous tau-positive threads in the white matter inclusions for CBD). Our findings support the feasibility of deep learning-based approaches for the classification of neurodegenerative disorders on WSIs. Further investigation of this method, focusing on clinicopathologic correlations, is warranted.

Lenvatinib inhibits the growth of gastric cancer patient-derived xenografts generated from a heterogeneous population.

BACKGROUND: Lenvatinib is a multitargeted tyrosine kinase inhibitor that is being tested in combination with immune checkpoint inhibitors to treat advanced gastric cancer; however, little data exists regarding the efficacy of lenvatinib monotherapy. Patient-derived xenografts (PDX) are established by engrafting human tumors into immunodeficient mice. The generation of PDXs may be hampered by growth of lymphomas. In this study, we compared the use of mice with different degrees of immunodeficiency to establish PDXs from a diverse cohort of Western gastric cancer patients. We then tested the efficacy of lenvatinib in this system. METHODS: PDXs were established by implanting gastric cancer tissue into NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) or Foxn1nu (nude) mice. Tumors from multiple passages from each PDX line were compared histologically and transcriptomically. PDX-bearing mice were randomized to receive the drug delivery vehicle or lenvatinib. After 21 days, the percent tumor volume change (%Δvtumor) was calculated. RESULTS: 23 PDX models were established from Black, non-Hispanic White, Hispanic, and Asian gastric cancer patients. The engraftment rate was 17% (23/139). Tumors implanted into NSG (16%; 18/115) and nude (21%; 5/24) mice had a similar engraftment rate. The rate of lymphoma formation in nude mice (0%; 0/24) was lower than in NSG mice (20%; 23/115; p < 0.05). PDXs derived using both strains maintained histologic and gene expression profiles across passages. Lenvatinib treatment (mean %Δvtumor: -33%) significantly reduced tumor growth as compared to vehicle treatment (mean %Δvtumor: 190%; p < 0.0001). CONCLUSIONS: Nude mice are a superior platform than NSG mice for generating PDXs from gastric cancer patients. Lenvatinib showed promising antitumor activity in PDXs established from a diverse Western patient population and warrants further investigation in gastric cancer.

Targeting renal cell carcinoma with a HIF-2 antagonist.

Clear cell renal cell carcinoma (ccRCC) is characterized by inactivation of the von Hippel-Lindau tumour suppressor gene (VHL). Because no other gene is mutated as frequently in ccRCC and VHL mutations are truncal, VHL inactivation is regarded as the governing event. VHL loss activates the HIF-2 transcription factor, and constitutive HIF-2 activity restores tumorigenesis in VHL-reconstituted ccRCC cells. HIF-2 has been implicated in angiogenesis and multiple other processes, but angiogenesis is the main target of drugs such as the tyrosine kinase inhibitor sunitinib. HIF-2 has been regarded as undruggable. Here we use a tumourgraft/patient-derived xenograft platform to evaluate PT2399, a selective HIF-2 antagonist that was identified using a structure-based design approach. PT2399 dissociated HIF-2 (an obligatory heterodimer of HIF-2α-HIF-1ß) in human ccRCC cells and suppressed tumorigenesis in 56% (10 out of 18) of such lines. PT2399 had greater activity than sunitinib, was active in sunitinib-progressing tumours, and was better tolerated. Unexpectedly, some VHL-mutant ccRCCs were resistant to PT2399. Resistance occurred despite HIF-2 dissociation in tumours and evidence of Hif-2 inhibition in the mouse, as determined by suppression of circulating erythropoietin, a HIF-2 target and possible pharmacodynamic marker. We identified a HIF-2-dependent gene signature in sensitive tumours. Gene expression was largely unaffected by PT2399 in resistant tumours, illustrating the specificity of the drug. Sensitive tumours exhibited a distinguishing gene expression signature and generally higher levels of HIF-2α. Prolonged PT2399 treatment led to resistance. We identified binding site and second site suppressor mutations in HIF-2α and HIF-1ß, respectively. Both mutations preserved HIF-2 dimers despite treatment with PT2399. Finally, an extensively pretreated patient whose tumour had given rise to a sensitive tumourgraft showed disease control for more than 11 months when treated with a close analogue of PT2399, PT2385. We validate HIF-2 as a target in ccRCC, show that some ccRCCs are HIF-2 independent, and set the stage for biomarker-driven clinical trials.

SH3BP4 is a negative regulator of amino acid-Rag GTPase-mTORC1 signaling.

Amino acids stimulate cell growth and suppress autophagy through activation of mTORC1. The activation of mTORC1 by amino acids is mediated by Rag guanosine triphosphatase (GTPase) heterodimers on the lysosome. The molecular mechanism by which amino acids regulate the Rag GTPase heterodimers remains to be elucidated. Here, we identify SH3 domain-binding protein 4 (SH3BP4) as a binding protein and a negative regulator of Rag GTPase complex. SH3BP4 binds to the inactive Rag GTPase complex through its Src homology 3 (SH3) domain under conditions of amino acid starvation and inhibits the formation of active Rag GTPase complex. As a consequence, the binding abrogates the interaction of mTORC1 with Rag GTPase complex and the recruitment of mTORC1 to the lysosome, thus inhibiting amino acid-induced mTORC1 activation and cell growth and promoting autophagy. These results demonstrate that SH3BP4 is a negative regulator of the Rag GTPase complex and amino acid-dependent mTORC1 signaling.

Transfer learning across ontologies for phenome-genome association prediction.

Motivation: To better predict and analyze gene associations with the collection of phenotypes organized in a phenotype ontology, it is crucial to effectively model the hierarchical structure among the phenotypes in the ontology and leverage the sparse known associations with additional training information. In this paper, we first introduce Dual Label Propagation (DLP) to impose consistent associations with the entire phenotype paths in predicting phenotype-gene associations in Human Phenotype Ontology (HPO). DLP is then used as the base model in a transfer learning framework (tlDLP) to incorporate functional annotations in Gene Ontology (GO). By simultaneously reconstructing GO term-gene associations and HPO phenotype-gene associations for all the genes in a protein-protein interaction network, tlDLP benefits from the enriched training associations indirectly through relation with GO terms. Results: In the experiments to predict the associations between human genes and phenotypes in HPO based on human protein-protein interaction network, both DLP and tlDLP improved the prediction of gene associations with phenotype paths in HPO in cross-validation and the prediction of the most recent associations added after the snapshot of the training data. Moreover, the transfer learning through GO term-gene associations significantly improved association predictions for the phenotypes with no more specific known associations by a large margin. Examples are also shown to demonstrate how phenotype paths in phenotype ontology and transfer learning with gene ontology can improve the predictions. Availability and Implementation: Source code is available at http://compbio.cs.umn.edu/onto phenome . Contact: kuang@cs.umn.com. Supplementary information: Supplementary data are available at Bioinformatics online.

An integrative somatic mutation analysis to identify pathways linked with survival outcomes across 19 cancer types.

MOTIVATION: Identification of altered pathways that are clinically relevant across human cancers is a key challenge in cancer genomics. Precise identification and understanding of these altered pathways may provide novel insights into patient stratification, therapeutic strategies and the development of new drugs. However, a challenge remains in accurately identifying pathways altered by somatic mutations across human cancers, due to the diverse mutation spectrum. We developed an innovative approach to integrate somatic mutation data with gene networks and pathways, in order to identify pathways altered by somatic mutations across cancers. RESULTS: We applied our approach to The Cancer Genome Atlas (TCGA) dataset of somatic mutations in 4790 cancer patients with 19 different types of tumors. Our analysis identified cancer-type-specific altered pathways enriched with known cancer-relevant genes and targets of currently available drugs. To investigate the clinical significance of these altered pathways, we performed consensus clustering for patient stratification using member genes in the altered pathways coupled with gene expression datasets from 4870 patients from TCGA, and multiple independent cohorts confirmed that the altered pathways could be used to stratify patients into subgroups with significantly different clinical outcomes. Of particular significance, certain patient subpopulations with poor prognosis were identified because they had specific altered pathways for which there are available targeted therapies. These findings could be used to tailor and intensify therapy in these patients, for whom current therapy is suboptimal. AVAILABILITY AND IMPLEMENTATION: The code is available at: http://www.taehyunlab.org CONTACT: jhcheong@yuhs.ac or taehyun.hwang@utsouthwestern.edu or taehyun.cs@gmail.com SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

TALEN-engineered AR gene rearrangements reveal endocrine uncoupling of androgen receptor in prostate cancer.

Androgen receptor (AR) target genes direct development and survival of the prostate epithelial lineage, including prostate cancer (PCa). Thus, endocrine therapies that inhibit the AR ligand-binding domain (LBD) are effective in treating PCa. AR transcriptional reactivation is central to resistance, as evidenced by the efficacy of AR retargeting in castration-resistant PCa (CRPC) with next-generation endocrine therapies abiraterone and enzalutamide. However, resistance to abiraterone and enzalutamide limits this efficacy in most men, and PCa remains the second-leading cause of male cancer deaths. Here we show that AR gene rearrangements in CRPC tissues underlie a completely androgen-independent, yet AR-dependent, resistance mechanism. We discovered intragenic AR gene rearrangements in CRPC tissues, which we modeled using transcription activator-like effector nuclease (TALEN)-mediated genome engineering. This modeling revealed that these AR gene rearrangements blocked full-length AR synthesis, but promoted expression of truncated AR variant proteins lacking the AR ligand-binding domain. Furthermore, these AR variant proteins maintained the constitutive activity of the AR transcriptional program and a CRPC growth phenotype independent of full-length AR or androgens. These findings demonstrate that AR gene rearrangements are a unique resistance mechanism by which AR transcriptional activity can be uncoupled from endocrine regulation in CRPC.

SHEAR: sample heterogeneity estimation and assembly by reference.

BACKGROUND: Personal genome assembly is a critical process when studying tumor genomes and other highly divergent sequences. The accuracy of downstream analyses, such as RNA-seq and ChIP-seq, can be greatly enhanced by using personal genomic sequences rather than standard references. Unfortunately, reads sequenced from these types of samples often have a heterogeneous mix of various subpopulations with different variants, making assembly extremely difficult using existing assembly tools. To address these challenges, we developed SHEAR (Sample Heterogeneity Estimation and Assembly by Reference; http://vk.cs.umn.edu/SHEAR), a tool that predicts SVs, accounts for heterogeneous variants by estimating their representative percentages, and generates personal genomic sequences to be used for downstream analysis. RESULTS: By making use of structural variant detection algorithms, SHEAR offers improved performance in the form of a stronger ability to handle difficult structural variant types and better computational efficiency. We compare against the lead competing approach using a variety of simulated scenarios as well as real tumor cell line data with known heterogeneous variants. SHEAR is shown to successfully estimate heterogeneity percentages in both cases, and demonstrates an improved efficiency and better ability to handle tandem duplications. CONCLUSION: SHEAR allows for accurate and efficient SV detection and personal genomic sequence generation. It is also able to account for heterogeneous sequencing samples, such as from tumor tissue, by estimating the subpopulation percentage for each heterogeneous variant.

The 3D Revolution in Cancer Discovery.

SUMMARY: The transition from 2D to 3D spatial profiling marks a revolutionary era in cancer research, offering unprecedented potential to enhance cancer diagnosis and treatment. This commentary outlines the experimental and computational advancements and challenges in 3D spatial molecular profiling, underscoring the innovation needed in imaging tools, software, artificial intelligence, and machine learning to overcome implementation hurdles and harness the full potential of 3D analysis in the field.

Integrative analysis of H&E and IHC identifies prognostic immune subtypes in HPV related oropharyngeal cancer.

BACKGROUND: Deep learning techniques excel at identifying tumor-infiltrating lymphocytes (TILs) and immune phenotypes in hematoxylin and eosin (H&E)-stained slides. However, their ability to elucidate detailed functional characteristics of diverse cellular phenotypes within tumor immune microenvironment (TME) is limited. We aimed to enhance our understanding of cellular composition and functional characteristics across TME regions and improve patient stratification by integrating H&E with adjacent immunohistochemistry (IHC) images. METHODS: A retrospective study was conducted on patients with Human Papillomavirus-positive oropharyngeal squamous cell carcinoma (OPSCC). Using paired H&E and IHC slides for 11 proteins, a deep learning pipeline was used to quantify tumor, stroma, and TILs in the TME. Patients were classified into immune inflamed (IN), immune excluded (IE), or immune desert (ID) phenotypes. By registering the IHC and H&E slides, we integrated IHC data to capture protein expression in the corresponding tumor regions. We further stratified patients into specific immune subtypes, such as IN, with increased or reduced CD8+ cells, based on the abundance of these proteins. This characterization provided functional insight into the H&E-based subtypes. RESULTS: Analysis of 88 primary tumors and 70 involved lymph node tissue images reveals an improved prognosis in patients classified as IN in primary tumors with high CD8 and low CD163 expression (p = 0.007). Multivariate Cox regression analysis confirms a significantly better prognosis for these subtypes. CONCLUSIONS: Integrating H&E and IHC data enhances the functional characterization of immune phenotypes of the TME with biological interpretability, and improves patient stratification in HPV( + ) OPSCC.

In this study, we investigated whether differences in the immune cell population surrounding head and neck cancers impact disease progression. We used advanced computer programs to analyze tissue samples from tumors and nearby lymph nodes, a part of the immune system. These tumor and lymph node samples were stained to show the structure of the tissue and to identify the different types of immune cells present. We grouped patients into different categories based on differences in their immune cells. We found that patients with certain patterns of immune cells tended to have better outcomes. This method could help doctors predict how well patients will respond to treatments.

Large-scale integrative network-based analysis identifies common pathways disrupted by copy number alterations across cancers.

BACKGROUND: Many large-scale studies analyzed high-throughput genomic data to identify altered pathways essential to the development and progression of specific types of cancer. However, no previous study has been extended to provide a comprehensive analysis of pathways disrupted by copy number alterations across different human cancers. Towards this goal, we propose a network-based method to integrate copy number alteration data with human protein-protein interaction networks and pathway databases to identify pathways that are commonly disrupted in many different types of cancer. RESULTS: We applied our approach to a data set of 2,172 cancer patients across 16 different types of cancers, and discovered a set of commonly disrupted pathways, which are likely essential for tumor formation in majority of the cancers. We also identified pathways that are only disrupted in specific cancer types, providing molecular markers for different human cancers. Analysis with independent microarray gene expression datasets confirms that the commonly disrupted pathways can be used to identify patient subgroups with significantly different survival outcomes. We also provide a network view of disrupted pathways to explain how copy number alterations affect pathways that regulate cell growth, cycle, and differentiation for tumorigenesis. CONCLUSIONS: In this work, we demonstrated that the network-based integrative analysis can help to identify pathways disrupted by copy number alterations across 16 types of human cancers, which are not readily identifiable by conventional overrepresentation-based and other pathway-based methods. All the results and source code are available at http://compbio.cs.umn.edu/NetPathID/.

Machine Learning Algorithm for Efficient Design of Separated Buffer Super-Junction IGBT.

An improved structure for an Insulated Gate Bipolar Transistor (IGBT) with a separated buffer layer is presented in order to improve the trade-off between the turn-off loss (Eoff) and on-state voltage (Von). However, it is difficult to set efficient parameters due to the increase in the new buffer doping concentration variable. Therefore, a machine learning (ML) algorithm is proposed as a solution. Compared to the conventional Technology Computer-Aided Design (TCAD) simulation tool, it is demonstrated that incorporating the ML algorithm into the device analysis could make it possible to achieve high accuracy and significantly shorten the simulation time. Specifically, utilizing the ML algorithm could achieve coefficients of determination (R2) of Von and Eoff of 0.995 and 0.968, respectively. In addition, it enables the optimized design to fit the target characteristics. In this study, the structure proposed for the trade-off improvement was targeted to obtain the minimum Eoff at the same Von, especially by adjusting the concentration of the separated buffer. We could improve Eoff by 36.2% by optimizing the structure, which was expected to be improved by 24.7% using the ML approach. In another way, it is possible to inversely design four types of structures with characteristics close to the target characteristics (Eoff = 1.64 µJ, Von = 1.38 V). The proposed method of incorporating machine learning into device analysis is expected to be very strategic, especially for power electronics analysis (where the transistor size is comparatively large and requires significant computation). In summary, we improved the trade-off using a separated buffer, and ML enabled optimization and a more precise design, as well as reverse engineering.

ACTA2 Expression Predicts Survival and Is Associated with Response to Immune Checkpoint Inhibitors in Gastric Cancer.

PURPOSE: We sought to identify biomarkers that predict overall survival (OS) and response to immune checkpoint inhibitors (ICI) for patients with gastric cancer. EXPERIMENTAL DESIGN: This was a retrospective study of multiple independent cohorts of patients with gastric cancer. The association between tumor ACTA2 expression and OS and ICI response were determined in patients whose tumors were analyzed with bulk mRNA sequencing. Single-cell RNA sequencing (scRNA-seq) and digital spatial profiling data were used to compare tumors from patients with gastric cancer who did and did not respond to ICI. RESULTS: Increasing tumor ACTA2 expression was independently associated with worse OS in a 567-patient discovery cohort [HR, 1.28 per unit increase; 95% confidence interval (CI), 1.02-1.62]. This finding was validated in three independent cohorts (n = 974; HR, 1.52 per unit increase; 95% CI, 1.34-1.73). Of the 108 patients treated with ICI, 56% of patients with low ACTA2 expression responded to ICI versus 25% of patients with high ACTA2 expression (P = 0.004). In an analysis of a publicly available scRNA-seq dataset of 5 microsatellite instability-high patients treated with ICI, the patient who responded to ICI had lower tumor stromal ACTA2 expression than the 4 nonresponders. Digital spatial profiling of tumor samples from 4 ICI responders and 5 ICI nonresponders revealed that responders may have lower ACTA2 expression in α-SMA-positive cancer-associated fibroblasts (CAF) than nonresponders (median: 5.00 vs. 5.50). CONCLUSIONS: ACTA2 expression is associated with survival and ICI response in patients with gastric cancer. ACTA2 expression in CAFs, but not in other cellular compartments, appears to be associated with ICI response.

Spatial heterogeneity and organization of tumor mutation burden with immune infiltrates within tumors based on whole slide images correlated with patient survival in bladder cancer.

Background: High tumor mutation burden (TMB-H) could result in an increased number of neoepitopes from somatic mutations expressed by a patient's own tumor cell which can be recognized and targeted by neighboring tumor-infiltrating lymphocytes (TILs). Deeper understanding of spatial heterogeneity and organization of tumor cells and their neighboring immune infiltrates within tumors could provide new insights into tumor progression and treatment response. Methods: Here we first developed computational approaches using whole slide images (WSIs) to predict bladder cancer patients' TMB status and TILs across tumor regions, and then investigate spatial heterogeneity and organization of regions harboring TMB-H tumor cells and TILs within tumors, as well as their prognostic utility. Results: In experiments using WSIs from The Cancer Genome Atlas (TCGA) bladder cancer (BLCA), our findings show that computational pathology can reliably predict patient-level TMB status and delineate spatial TMB heterogeneity and co-organization with TILs. TMB-H patients with low spatial heterogeneity enriched with high TILs show improved overall survival. Conclusions: Computational approaches using WSIs have the potential to provide rapid and cost-effective TMB testing and TILs detection. Survival analysis illuminates potential clinical utility of spatial heterogeneity and co-organization of TMB and TILs as a prognostic biomarker in BLCA which warrants further validation in future studies.

Multi-omics reveals microbiome, host gene expression, and immune landscape in gastric carcinogenesis.

To date, there has been no multi-omic analysis characterizing the intricate relationships between the intragastric microbiome and gastric mucosal gene expression in gastric carcinogenesis. Using multi-omic approaches, we provide a comprehensive view of the connections between the microbiome and host gene expression in distinct stages of gastric carcinogenesis (i.e., healthy, gastritis, cancer). Our integrative analysis uncovers various associations specific to disease states. For example, uniquely in gastritis, Helicobacteraceae is highly correlated with the expression of FAM3D, which has been previously implicated in gastrointestinal inflammation. In addition, in gastric cancer but not in adjacent gastritis, Lachnospiraceae is highly correlated with the expression of UBD, which regulates mitosis and cell cycle time. Furthermore, lower abundances of B cell signatures in gastric cancer compared to gastritis may suggest a previously unidentified immune evasion process in gastric carcinogenesis. Our study provides the most comprehensive description of microbial, host transcriptomic, and immune cell factors of the gastric carcinogenesis pathway.

Spatial analysis of tumor-infiltrating lymphocytes in histological sections using deep learning techniques predicts survival in colorectal carcinoma.

This study aimed to explore the prognostic impact of spatial distribution of tumor-infiltrating lymphocytes (TILs) quantified by deep learning (DL) approaches based on digitalized whole-slide images stained with hematoxylin and eosin in patients with colorectal cancer (CRC). The prognostic impact of spatial distributions of TILs in patients with CRC was explored in the Yonsei cohort (n = 180) and validated in The Cancer Genome Atlas (TCGA) cohort (n = 268). Two experienced pathologists manually measured TILs at the most invasive margin (IM) as 0-3 by the Klintrup-Mäkinen (KM) grading method and this was compared to DL approaches. Inter-rater agreement for TILs was measured using Cohen's kappa coefficient. On multivariate analysis of spatial TIL features derived by DL approaches and clinicopathological variables including tumor stage, microsatellite instability, and KRAS mutation, TIL densities within 200 µm of the IM (f_im200) remained the most significant prognostic factor for progression-free survival (PFS) (hazard ratio [HR] 0.004 [95% confidence interval, CI, 0.0001-0.15], p = 0.0028) in the Yonsei cohort. On multivariate analysis using the TCGA dataset, f_im200 retained prognostic significance for PFS (HR 0.031 [95% CI 0.001-0.645], p = 0.024). Inter-rater agreement of manual KM grading was insignificant in the Yonsei (κ = 0.109) and the TCGA (κ = 0.121) cohorts. The survival analysis based on KM grading showed statistically significant different PFS in the TCGA cohort, but not the Yonsei cohort. Automatic quantification of TILs at the IM based on DL approaches shows prognostic utility to predict PFS, and could provide robust and reproducible TIL density measurement in patients with CRC.