The Children’s Tumor Foundation hosted its fourth Hackathon, the HACK4NF, from October 14 – November 7, 2022. The goal of this virtual hackathon was to advance research for all types of NF by utilizing genomic datasets to develop drug targets, predict NF variants and correctly classify NF tumors. More than 300 people from diverse fields participated in the HACK4NF, including those with expertise in computational biology, predictive AI, genomics, data analytics, and medical research.
The Hack4NF Project Presentations were live-streamed on Tuesday, November 8, 2022. Each participant shared a three-minute presentation followed by a short Q&A with each team. These presentations may now be viewed on the CTF YouTube channel at this link.
“We are very happy with the quality of the submissions we had in the HACK 4NF 2022 edition. The focus on data and the specifics of the challenges helped us attract a highly-skilled audience that used machine learning techniques to create complex analyses and working prototypes with the potential to help researchers with new computational tools,” said Salvatore La Rosa, PhD, Chief Scientific Officer of the Children’s Tumor Foundation. “We also saw very strong incubation proposals with two winners and other exciting ideas, that we hope to move forward.”
Following their initial presentations, for three weeks participants formed teams, analyzed datasets, engaged with patients, and received mentorship from 24 experts in the field. At the conclusion of the hackathon, there were 21 project pitches, 5 winning teams, and more than $18,500 in prizes awarded. While approximately 46 participants reported having an NF connection (family member, friend, caregiver, or patient), over 75% (266) did not have an NF connection, opening the door and inviting new minds to lend their brain power in solving research questions around NF.
“As a mom of a 5-year-old with NF, I was overwhelmed by the outpouring of talent, expertise and enthusiasm participants in the Hackathon displayed,” said Danielle Bonadies, HACK4NF mentor. “With an incredibly tight timeline, teams developed and submitted extraordinary projects that have the potential to push research and ideas in the NF community forward by leaps and bounds. It gives me a renewed sense of community and hope to know that so many are willing to donate their talents for the benefit of others.”
Teams were asked to create projects in one of three challenge categories:
- Challenge 1 – GENIE-NF tumor identification and classification challenge
- Use the provided Genomics Evidence Neoplasia Information Exchange (GENIE) datasets provided to develop a new framework that accurately uses genomic data to classify tumor samples for neurofibromatosis-related tumors.
- Challenge 2 – Devising in silico strategies to prioritize likely pathogenic NF1 germline variants
- Develop a strategy to score the pathogenicity of individual NF1 variants. Your solution may combine features of the variant, predictions and findings from external data sources, AlphaFold protein structure predictions, literature-mining, or other evidence sources to assign a severity score to any NF1 variant.
- Challenge 3 – In silico drug target screening for NF
- Use variant, gene expression, or drug screening data from different sources and apply already-published or new methods to predict drug targets that could be helpful in treating manifestations of NF1, NF2, or schwannomatosis.
Once final projects were submitted, judges scored the projects by weighing their impact, innovation, technical execution, and presentation.
“It was a great honor being involved in the 2022 HACK4NF again this year! I was inspired by the hard work and creativity of each of the teams,” said HACK4NF judge Carlos Romo, PhD, Director of Clinical Research, Neurofibromatosis Therapeutic Acceleration Program, Johns Hopkins University School of Medicine. “The challenges were difficult to tackle, but the degree of innovation in the proposals made selecting the winning teams difficult. I am grateful for the dedication of the participants and organizers, and confident the projects will yield great fruits for the NF community.”
Winners were announced during the 2022 Hack4NF Award Presentations, live-streamed on Wednesday, November 16. Awards were presented to winning teams for 1st Place in Challenge, “Best of” Awards, and Incubation Prizes. Watch the award presentations in the video below, or scroll down for a list of the 2022 winners and their project descriptions.
2022 Hack4NF Winners
Winning Project: Next GeNLP
- 1st Place, Challenge Category 1: GENIE-NF tumor identification and classification challenge
- Awarded $1,500 Prize for 1st Place in Challenge
- Also awarded $1,000 Prize for Best Use of PMP
- Team Members: Gabriel Altay, Hari (Hariprasad) Donthi, Karthika Kavarmar, William Xie
- Project Description: Finding structure in tumor genetic features by embedding, clustering, and mapping.
- The Issue: Our challenge is to identify the structure in the genetic profiles of NF tumors. We have a large collection of mutation and copy number alteration data from GENIE, but the NF related portion (identified by NF-relevant genes or tumors that are common in patients with NF) is relatively small.
- The Solution: We created tumor sample embeddings by applying natural language processing techniques to genomic data. Our embeddings can capture variant and copy number alteration data for each sample in a continuous high dimensional space. We explored the structure of this tumor sample embedding space using an interactive 3-D visualization tool called tensorflow projector. We also showed that these embeddings can be used to predict GENIE cancer types although the performance was poor for NF related cancer types. We made use of the AHA Precision Medicine Platform for computing resources.
Winning Project: DITTO4NF
- 1st Place, Challenge Category 2: Devising in silico strategies to prioritize likely pathogenic NF1 germline variants
- Awarded $1,500 Prize for 1st Place in Challenge
- Also awarded $1,000 Prize for Best Use of PMP and $5,000 Incubation Prize
- Team Members: Christian Fay, Gurpreet Kaur, Tarun Karthik Kumar Mamidi, Ravikul Rao, Elizabeth Worthey
- Project Description: Our explainable machine learning tool, DITTO, prioritizing germline variants for NF1.
- The Issue: Neurofibromatosis type 1 (NF1) is a genetic condition that affects various human systems such as the skin, the skeleton and the part of the nervous system outside the brain and spinal cord peripheral nervous system. Identifying the pathogenicity of variants still remains a challenge for precision diagnosis and prognosis. The goal of this challenge is to devise in silico strategies to prioritize likely pathogenic NF1 germline variants.
- The Solution: We will use publicly available datasets and several predictors ranging from genome level to protein 2D and 3D structure (for stability and interactions) to predict variant deleteriousness. We will then map these deleterious variants to individual's transcriptome and phenotype information to support variant classification and prioritization providing definitive diagnosis for patients with NF1.
Winning Project: NF1 Drug Targeting
- 1st Place, Challenge Category 3: In silico drug target screening for NF
- Awarded $1,500 Prize for 1st Place in Challenge
- Also awarded $5,000 Incubation Prize
- Team Members: Richard Conroy, Chang In Moon, Aastha Naik, Matthew Zamora, Paul Zamora
- Project Description: Challenging NF1 through thoughtful drug screening
- The Issue: NF1 has no commonly accepted drug for adults. Providing a list of possible drugs matched with genetic variants, biochemical assays, and clinical presentation would support thoughtful choices for further investigation. There may not be a single best drug treatment for each case of NF1 as the pathology is on a spectrum, and symptoms vary.
- The Solution: We filter small molecule data from quantified High Throughput Screening using statistical interpretations. We identify trends in responsive compounds, use network analysis of pathways, and understand mechanisms of action to find patterns and relationships that would inform treatment options. We then demonstrate a more detailed application of drug sensitivity screening for malignant forms of NF1.
Winning Project: Artificial Intelligence for NF
- Awarded $1,000 Prize for Best Documentation
- Submitted for Challenge Category 1: GENIE-NF tumor identification and classification challenge
- Team Members: Muhammad Alaa Alwattar, Harshal Chorya, Lucas Pastur Romay, Jeffy Joseph Vinohar
- Project Description: Machine Learning and Natural Language Processing for neurofibromatosis-related tumors
- The Issue: The Identification and pathogenic classification of NF gene variant tumors among other types in a patient sample pool. It is difficult to classify tumor samples for neurofibromatosis-related tumors from other types of tumors.
- The Solution:
- Use Machine Learning (ML) to find each cancer type's most relevant gene mutations.
- Apply Natural Language Processing (NLP) techniques to cluster related genes.
- We developed a web application to interact with the ML models results and NLP visualizations
Winning Project: Drug Discovery for NF1
- Awarded $1,000 Prize for Best Use of Data
- Submitted for Challenge Category 3: In silico drug target screening for NF
- Team Members: Yunguang Sun, Kathy Sun
- Project Description: Reversing the detrimental gene expression in pNF by integrated drug screen mining
- The Issue: Prioritizing candidates in a drug screen:
- Potential drug response bias from screening data
- Library of Integrated Network-Based Cellular Signatures (LINCS) database
- Reverse the highly expressed gene expression in pNF to inhibit tumor growth
- The Solution: We utilized publicly available gene expression and small compound screening datasets to rank and prioritize the potential compounds to treat pNF1.
This year’s strategic supporters for the hackathon were the AACR Project GENIE, OpenCRAVAT, the Precision Medicine Platform provided by American Heart Association, and Sage Bionetworks.
The 2022 HACK4NF was presented by Alexion AstraZeneca Rare Disease, Recursion, SpringWorks Therapeutics, and NTAP (Neurofibromatosis Therapy Acceleration Program) with cloud support from the American Heart Association’s Precision Medicine Platform, and datasets provided by Sage Bionetworks, NF Data Portal, NF Open Science Initiative, OpenCRAVAT, Dr. Rick Van Minkelen, and the American Association of Cancer Research’s Project GENIE.
