2:00 pm – 2:10 pm

Welcome – Ninth Computational Approaches for Cancer Workshop (CAFCW)

Sally R. Ellingson, PhD, University of Kentucky

Eric A. Stahlberg, PhD, Frederick National Laboratory for Cancer Research

2:10 pm – 2:40 pm

Overcoming the Challenges to Democratizing Precision Medicine: HPC Infrastructure, Health Equity Training Sets, Training a Diverse Workforce, and Mitigating Fears

Featured Speaker — James W. Lillard, PhD, MBA, Morehouse School of Medicine

Introduced by Eric A. Stahlberg, PhD, Frederick National Laboratory for Cancer Research

2:40 pm - 2:55 pm

Session 1         Patient Scale Computational Approaches for Cancer

AI/ML-Derived Whole-Genome Predictor Prospectively and Clinically Predicts Survival and Response to Treatment in Brain Cancer

Presenter: Orly Alter, PhD, University of Utah

Authors: Sri Priya Ponnapalli, PhD, University of Utah; Penelope Miron, PhD, Case Western Reserve University School of Medicine; Kristy L. S. Miskimen, PhD, Case Western Reserve University School of Medicine; Kristin A. Waite, PhD, National Cancer Institute; Nadiya Sosonkina, PhD, HudsonAlpha Clinical Services Lab LLC; Sara E. Coppens, B.Sc., The Abigail Wexner Research Institute at Nationwide Children’s Hospital; Anthony C. Bryan, PhD, The Abigail Wexner Research Institute at Nationwide Children’s Hospital; Estevan P. Kiernan, M.Sc., Illumina, Inc.; Huanming Yang, PhD, Beijing Genomics Institute (BGI) -Shenzhen; Jay Bowen, M.Sc., The Abigail Wexner Research Institute at Nationwide Children’s Hospital; Ghunwa A. Nakouzi, PhD, HudsonAlpha Clinical Services Lab LLC; Jill S. Barnholtz-Sloan, PhD, National Cancer Institute; Andrew E. Sloan, MD, Case Western Reserve University School of Medicine; Tiffany R. Hodges, MD, Case Western Reserve University School of Medicine; Orly Alter, PhD, University of Utah.

Abstract: Cancer is complex, with contributing factors distributed across the entire genome affecting every aspect of the disease. But typical artificial intelligence and machine learning (AI/ML) would require 3B-patient training sets to generate predictive models from the whole 3B-nucleotide genome. As a result, tests remain limited to one to a few hundred genes. Prediction continues to rely mostly on such factors as a tumor’s grade and the patient’s age. And the understanding and management of cancer continue to involve guesswork.

A genome-wide pattern in glioblastoma brain cancer tumors was experimentally validated in a retrospective clinical trial as the most accurate and precise predictor of life expectancy and response to standard of care [1]. Applicable to the general population, this predictor, the first to encompass the whole genome, and predictors in lung, nerve, ovarian, and uterine cancers, were mathematically (re)discovered and computationally (re)validated in open-source datasets from as few as 50–100 patients by using our AI/ML [2,3]. Data-agnostic, our algorithms, multi-tensor comparative spectral decompositions, extend the mathematics that underlies quantum mechanics to overcome typical AI/ML obstacles by not requiring large amounts of data, balanced data, or feature engineering. All other attempts to connect a glioblastoma patient’s outcome with the tumor’s DNA copy numbers failed. For 70 years, the best indicator has been age. At 75–95% accuracy, our predictor is more accurate than and independent of age and all other indicators. Platform- and reference genome-agnostic, the predictor’s >99% precision is greater than the community consensus of <70% reproducibility based upon one to a few hundred genes. It describes mechanisms fortransformation and identifies drug targets and combinations of targets to sensitize tumors to treatment.

Now, in follow-up results from the trial we, first, show correct prospective prediction of the outcome of the five of the 79 patients who were alive four years earlier, at the time of first results. Two patients, who were predicted to have shorter survival, lived less than five years from diagnosis, whereas of the three patients predicted to have longer survival, one lived more than five, and the remaining two are alive >11.5, years from diagnosis. Second, we demonstrate 100%-precise clinical prediction for 59 of the 79 patients with remaining tumor DNA by using whole-genome sequencing in a regulated laboratory. Third, we establish that the risk that a tumor’s whole genome confers upon outcome, as is reflected by the predictor, is surpassed only by the patient’s access to radiotherapy.

This is a proof of principle that our AI/ML is uniquely suited for personalized medicine. This also demonstrates that the inclusion of complete genomes, and the normal diversity within, is, beyond fair AI/ML, a scientific, engineering, and medical necessity, because a patient’s survival and response to treatment are the outcome of their tumor’s whole genome. We conclude that our AI/ML-derived whole-genome predictors can take the guesswork out of cancer.

[1] Ponnapalli et al., APL Bioeng 4, 026106 (2020); https://doi.org/10.1063/1.5142559
[2] Bradley et al., APL Bioeng 3, 036104 (2019); https://doi.org/10.1063/1.5099268
[3] Alter et al., PNAS 100, 3351 (2003); https://doi.org/10.1073/pnas.053025810

CAFCW Announcements

Presenter: Eric A. Stahlberg, PhD, Frederick National Laboratory for Cancer Research

3:00 pm – 3:30 pm

Break – Virtual Poster Session Visits

Posters may be viewed starting November 12, 2023 

https:// https://cafcw23.virtualpostersession.org/

Panel – Diversity, Equity, and Inclusion – from Data to Workforce

Sally R. Ellingson, PhD, University of Kentucky, Moderator

Panelists:

Todd Burus, MAS, University of Kentucky

Sylvia Crivelli, PhD, Lawrence Berkeley National Laboratory

James W. Lillard, PhD, MBA, Morehouse School of Medicine

Lavanya Vishwanatha, PhD, National Institutes of Health

4:00 pm – 4:15 pm

Automated Whole-Body Tumor Segmentation and Prognosis of Cancer on PET/CT

Presenter/Author: Kevin H. Leung, PhD, Johns Hopkins University

Abstract: Background: Cancer is the second leading cause of death in the United States [1]. Automatic characterization of malignant disease is an important clinical need to facilitate early detection and treatment of cancer [2]. Advances in machine learning (ML) and deep learning (DL) have shown significant promise for radiological and oncological applications [3]. Radiomic analysis extracts quantitative features from radiologic data about a cancerous tumor [4]. DL methods require large training datasets with sufficiently annotated images, which are difficult to obtain for radiological applications. The objective of this study was to develop a deep semi-supervised transfer learning approach for automated whole-body tumor segmentation and prognosis on positron emission tomography (PET)/computed tomography (CT) scans using limited annotations (Fig. 1a).

Methods: Five datasets consisting of 1,019 prostate, lung, melanoma, lymphoma, head and neck, and breast cancer patients with prostate-specific membrane antigen (PSMA) and fluorodeoxyglucose (FDG) PET/CT scans were used in this study (Table 1). A nnUnet backbone was cross-validated on the tumor segmentation task via a 5-fold cross-validation. Predicted segmentations were iteratively improved using radiomic analysis. Transfer learning generalized the segmentation task across PSMA and FDG PET/CT. Segmentation accuracy was evaluated on true positive rate (TPR), positive predictive value (PPV), Dice similarity coefficient (DSC), false discovery rate (FDR), true negative rate (TNR), and negative predictive value (NPV). Imaging measures quantifying molecular tumor burden and uptake were extracted from the predicted segmentations. A risk stratification model was developed for prostate cancer by combining the extracted imaging measures and was evaluated on follow-up prostate-specific antigen (PSA) levels. A risk stratification model was developed for head and neck cancer patients by combining imaging measures and American Joint Committee on Cancer (AJCC) staging and was evaluated via Kaplan-Meier survival analysis. A prognostic model was developed to predict pathological response of breast cancer patients toneoadjuvant chemotherapy using imaging measures from pre-therapy and post-therapy PET/CT scans. Prognostic models were evaluated on overall accuracy and area under the receiver operating characteristic (AUROC) curve. Statistically significant differences were inferred using a Wilcoxon rank-sum test.

Results: Accuracy metrics and illustrative examples of predicted tumor segmentations are shown in Table 2 and Fig. 1b. The risk stratification model yielded an overall accuracy of 0.83 and an AUROC of 0.86 in stratifying prostate cancer patients (Fig. 1c). Median follow-up PSA levels in the low-intermediate and high risk groups were 1.19 ng/mL and 53.20 ng/mL (P < 0.05). Head and neck cancer patients were stratified into low, intermediate, and high risk groups with significantly different Kaplan-Meier survival curves by the log-rank test (Fig. 1d). A prognostic model using imaging measures from pre-therapy scans predicted pathological complete response (pCR) in breast cancer patients with an accuracy of 0.72 and an AUROC of 0.72. The model using imaging measures from both pre-therapy and post-therapy scans predicted pCR in breast cancer patients with an accuracy of 0.84 and an AUROC of 0.76.

Conclusion: A deep semi-supervised transfer learning approach was developed and demonstrated accurate tumor segmentation, quantification, and prognosis on PET/CT of patients across six cancer types.

4:15 pm — 4:30 pm

Optimized Patient-Specific Catheter Placement for Convection-Enhanced Nanoparticle Delivery in Recurrent Glioblastoma

Presenter/Author: Chengyue Wu, PhD, University of Texas, Oden Institute

Abstract: Introduction: Glioblastoma multiforme (GBM) is the most common and deadliest of all primary brain cancers. One promising treatment strategy for patients with recurrent GBM is convection-enhanced delivery (CED) of Rhenium-186 (186Re)-nanoliposomes (RNL) to provide delivery of large, localized doses of radiation. The success of treatment by CED relies on proper catheter placement for therapy delivery to maximize tumor coverage and minimize the leakage to healthy tissue. In this project, we are developing an image-guided physics-based model to optimize catheter placement for RNL delivery on a patient-specific basis.

Methods: The mathematical model consists of 1) the steady-state flow field generated via the catheter infusion and the Darcy flow through the 3D brain domain, 2) the transport of RNL governed by an advection-diffusion equation, and 3) the point-spread function to transform the RNL distribution into the SPECT signal. Pre-delivery MRIs were used to assign patient-specific tissue geometries. Two scenarios were performed to personalize the model parameters: a) patient-specific calibration with longitudinal SPECT images monitoring RNL distributions, and b) population-based assignment with the leave-one-out cross-validation (LOOCV). The accuracy of model predictions was evaluated by the concordance correlation coefficients (CCC) between predicted and measured voxel-wise SPECT signals. Furthermore, in each patient, we used the image-guided model—with either calibrated or assigned parameters—to simulate RNL distributions for all possible locations of catheter tip(s), resulting in a ratio of the cumulative dose of RNL outside the tumor to that within the tumor, termed as “off-target ratio” (OTR). We minimized the OTR to optimize the placement of catheter(s) and compared OTRs obtained by the optimized and the original placements.

Results: Fifteen patients with recurrent GBM from a Phase I/II clinical trial of RNL were included in the study. For scenario a) with the patient-specific calibrated parameters, our model achieved median CCCs of 0.91, 0.87, and 0.82 for predicting RNL distributions at the mid-delivery, end-of-delivery, and 24 h post-delivery, respectively. For scenario b) with the LOOCV assigned parameters, our model achieved median CCCs of 0.89, 0.84, and 0.79 for predicting RNL distributions at the mid-delivery, end-of-delivery, and 24 h post-delivery, respectively. Compared to the original catheter placements, the optimized placements with the patient-specifically calibrated model achieved a median (range) of 34.56% (14.70% – 61.12%) reduction on OTR at the 24h post-delivery. Similarly, the optimized placements with the LOOCV assigned model achieved a 34.56% (13.30% – 56.62%) reduction on OTR at the 24h post-delivery. Furthermore, the optimization provides insights into whether a patient is a proper candidate for CED of RNL, and whether a reduction of catheter number is possible for thepatient.

Conclusion: Our image-guided model, with either patient-specific calibrated parameters or LOOCV assigned parameters, achieved high accuracy for predicting RNL distributions up to 24 h after the RNL delivery. The placement of catheter(s) optimized via our modeling substantially reduced the off-target ratio of RNL delivery. These results proved the potential of our image-guided modeling to guide patient-specific optimization of catheter placement for convection-enhanced delivery of radiolabeled liposomes.

Acknowledgments: NCI R01CA235800, U01CA253540, and R01CA260003. CPRIT RR160005.

4:30 pm – 4:45 pm

CAFCW23 Student Presentation*

Environmental Factors and Lung Cancer: A Predictive Spatial Approach

Presenter: Wenhuan Tan, BS

Authors: Wenhuan Tan, BS, Xiange Wang, BS, Silvia Crivelli, PhD, Xinlian Liu, PhD

Abstract: Lung cancer has witnessed a substantial increase in prevalence over the past few decades. While studies have established that the environment is the primary cause of most lung cancer cases, the development of lung cancer may be the result of the combined impact of multiple environmental factors.

Our objective is to investigate the relationship between lung cancer incidence and various physical ambient factors, including climatology, air quality, meteorological conditions, and soil vegetation. To address the issue of missing data on lung cancer cases at the county level in 2020, we use a Bayesian spatial and temporal modeling approach to mapping geographic variation in lung cancer mortality rates for subnational areas with R-INLA.

Our predictive model is constructed using multiple independent variables obtained from various satellite sources, covering the period from 1960 to 2016. Climate data such as heatwaves and extremetemperatures are from the National Oceanic and Atmospheric Administration (NOAA) and PRISM climate group (PRISM). Air quality indicators such as PM2.5, NO2, and ozone levels are sourced from NASA's Earth Data. Observational meteorological data, encompassing temperature, dew point, wind direction, wind speed, cloud cover, cloud layers, ceiling height, visibility, current weather, and precipitation amount, are obtained from the EPA's high-resolution gridded dataset. Soil vegetation and cropland data are acquired from the United States Department of Agriculture (USDA) using satellite imagery. Furthermore, we explore additional geophysical data available through the Google Earth Engine platform.

Our predictive model reveals an increasing positive association between multiple environmental factors and lung cancer incidence over the years. We apply a linear model with group fixed effects to 2012-2016 data, assessing lung cancer's relative risk and generating a 2017-2021 environmental vulnerability map. This work highlights AI and integrated data analysis' potential in interpreting and predicting complex health phenomena like lung cancer.

*The CAFCW Student Presentation is dedicated to the memory of Petrina Chong Hollingsworth, who, as a member of the CAFCW Organizing Committee, was passionate about engaging students and who helped create the Student Track for CAFCW20. Sadly, we lost Petrina in July 2023 due to a rare form of cancer. As we continue the CAFCW tradition for the student presentation, we honor her memory, her enthusiasm, and her devotion.

Session 2       

Molecular Scale Computational Approaches for Cancer

4:45 pm – 5:00 pm

Constructing a large-scale biomedical knowledge graph and its applications in drug discovery

Presenter/Author: Jinfeng Zhang, PhD

Abstract:  In the past few decades, the biomedical research community has acquired a wealth of knowledge, much of which is stored in scientific literature as unstructured text. Converting this text into structured form is crucial for developing new methodologies and applications that can fully utilize this knowledge. To achieve this goal, two basic problems must be addressed: named entity recognition (NER) and relation extraction (RE). NER involves identifying the concepts or entities in texts, such as diseases, genes/proteins, and chemical compounds. RE, on the other hand, aims to extract the relationships between these entities. The information extracted fromNER and RE can be used to create knowledge graphs, where nodes represent entities in the text and edges represent their relationships. This presentation will discuss our team's work on the LitCoin NLP Challenge organized by NIH, for which we were awarded first place. Using pipelines developed for the challenge, we processed all PubMed articles and created a large-scale biomedical knowledge graph. The accuracy of this large-scale relation extraction isestimated using manual verification of a sample of the extracted data and found to be at the human annotation level. We also incorporated relation information from 40 public databases and relations inferred from publicly available genomics datasets. Our knowledge graph consists of over 11 million entities and more than 40 million relations. We have developed versatile query functions and knowledge discovery tools for accessing and mining structured data in the knowledge graph. Finally, we will discuss some drug discovery-relatedapplications enabled by this large-scale knowledge graph.

5:00 pm – 5:15 pm

Entropy-Based Regularization on Deep Learning Models for Anti-Cancer Drug Response Prediction

Presenter: Oleksandr Narykov, PhD

Authors: Oleksandr Narykov, PhD, Argonne National Laboratory (ANL), Data Science and Learning Division; Yitan Zhu, PhD, Argonne National Laboratory (ANL), Data Science and Learning Division; Thomas Brettin, MS, Argonne National Laboratory (ANL), Data Science and Learning Division; Yvonne Evrard, PhD, Frederick National Laboratory for Cancer Research; Alexander Partin, PhD, Argonne National Laboratory (ANL), Data Science and Learning Division; Maulik Shukla, MS, Argonne National Laboratory (ANL), Data Science and Learning Division; Priyanka Vasanthakumari, PhD, Argonne National Laboratory (ANL), Data Science and Learning Division; James Doroshow, MD, National Cancer Institute; Rick Stevens, Argonne National Laboratory.

Abstract: This work studies a particular setting for regression problems – tasks with complex combinatorial data space where samples can be divided into distinct groups. Anti-cancer drug response prediction is a perfect example of this setting, in which each sample includes cancer biological features and drug chemical information. Many existing works of pan-drug and pan-cancer response modeling treat different combinations of drugs and cancers as individual samples. A potential problem in these works is that a model may be heavily influenced and biased toward overrepresented drugs and cancers.

In the drug response prediction field, the performance of pan-cancer pan-drug models is commonly evaluated on a holdout test set through cross-validation (CV) using performance metrics like thecoefficient of determination (R2) and the mean squared error (MSE). However, drug response prediction can be viewed as a multi-objective optimization task, attempting to maximize the prediction performance over different drugs and cancers. We consider the performance for each drug as a separate objective and attempt to find a model on the corresponding Pareto front that provides balanced performances for all compounds. We propose adding an entropy-based regularization to the loss function for model training to reach this balanced state. The intuition behind it is straightforward – we minimize the MSE on all data points while encouraging the drug-specific model fitting error variability to stayas low as possible. We achieve this by grouping samples by their drug identity and computing the MSE for each group in the training batch. Then we calculate entropy over normalized group-specific losses. This value is plugged into the regularization term that incentivizes the loss function to maximize it. The maximum entropy can be achieved when the MSEs across all drugs take the same value.

We investigate the regularization effect on response modeling using a drug screening dataset of Cancer Cell Line Encyclopedia (CCLE) and a state-of-the-art drug response prediction model DeepTTA [1]. We consider two CV strategies for model evaluation – random split and drug-blind split. In a random split, the testing set can share both cell lines and drugs with the training set, while in a drug-blind split, a drug can not appear simultaneously in both the training and testing sets. We perform 10-fold CV analyses and evaluate the model performance using R2. For the random split, we see that adding the entropy-based regularization leads to a marginal improvement in prediction performance, which is 0.736 without regularization versus 0.746 with regularization and a p-value of 0.130 from the pairwise t-test. Importantly, we observe a substantial improvement in the more challenging setting of drug-blind split, where the prediction performance increases from -0.128 (without regularization) to 0.168 (with regularization) with a statistically significant p-value of 0.005 from pair-wise t-test.

1 Jiang, L., Jiang, C., Yu, X., Fu, R., Jin, S., and Liu, X.: ‘DeepTTA: a transformer-based model for predicting cancer drug response’, Briefings in Bioinformatics, 2022, 23, (3), pp. bbac100

5:15 pm – 5:30 pm

Scalable Lead Prediction with Transformers using HPC resources

Presenter: Archit Vasan, PhD, Argonne National Laboratory

Authors:  Archit Vasan, PhD, Argonne National Laboratory; Rick Stevens, Argonne National Laboratory; Arvind Ramanathan, PhD, Argonne National Laboratory; Venkatram Vishwanath, PhD, Argonne National Laboratory.

Abstract: A promising direction in cancer drug discovery is high-throughput screening of extensive compound datasets to identify advantageous properties, including their ability to interact with relevant biomolecules such as proteins. However, traditional structural approaches for assessing binding affinity, such as free energy methods or molecular docking, pose significant computational bottlenecks when dealing with such vast datasets. To address this, we have developed a docking surrogate called the SMILES transformer (ST), which learns molecular features from the SMILES representation of compounds and approximates their binding affinity. SMILES data is first tokenized using a well-established SMILES-pair tokenizer and fed into a BERT-like Transformer model to generate vector embeddings for each molecule, effectively capturing the essential information. These extracted embeddings are then fed into a regression model to predict the binding affinity. Leveraging the high-performance computing resources at Argonne National Lab, we devised a workflow to scale model training and inference across multiple supercomputing nodes. To evaluate the performance and accuracy of our workflow, we conducted experiments using molecular docking binding affinity data on multiple receptors, comparing ST with another state-of-the-art docking surrogate. Impressively, both surrogates yielded comparable val-r2 measurements of between 70 and 90%, affirming the capability of ST to learn molecular features directly from language-based data. Furthermore, one significant advantage of the ST approach is its notably faster tokenization preprocessing compared to the alternative method, which requires generating molecular descriptors using Mordred. Our workflow facilitated screening of ~ 3 billion compounds on 48 nodes of the Polaris supercomputer in approximately an hour. In summary, our approach presents an efficient means to screen extensive compound databases for potential molecular properties that could serve as lead compounds targeting cancer. Looking ahead, an important future direction for our workflow involves integrating de-novo drug design, enabling us to scale our efforts to explore the limits of synthesizable compounds within chemical space.

5:30 pm           

Workshop Concludes