{ "fetched_at": "2026-10-19T22:30:05Z", "source_url": "https://www.kaggle.com/competitions/genentech-404-challenge/data", "pages": [ { "name": "data-description", "content": "## Data Access\nAll participants are required to sign and adhere to the Data Usage Agreement for the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset on the LONI website [https://ida.loni.usc.edu/collaboration/access/appApply.jsp?project=ADNI](https://ida.loni.usc.edu/collaboration/access/appApply.jsp?project=ADNI). Participants must request access to the database, be aware that approval may take between 3-5 business days. Participants who already have approval to access ADNI through the LONI website, do not need to apply again. PARTICIPANTS SHOULD NOT DOWNLOAD THE COMPETITION DATA UNTIL THE ADNI DATA ACCESS REQUEST HAS BEEN APPROVED BY THE LONI WEBSITE ADMINISTRATORS. \n\nIn the Proposed Analysis section, Participants are encouraged to use the following statement:\n> I am requesting access to the ADNI database in order to participate in Genentech's 404 Challenge on missing data imputation. The goal of the 404 Challenge to develop flexible algorithms to deal with different patterns of missing data in tabular clinical datasets that include demographics, cognitive scores, and MR-derived volumetric measurements.\n\n## Files\nThe training set (dev_set.csv) is a complete dataset with no missing values. Three copies of dev_set.csv are created, dev_1.csv, dev_2.csv, dev_3.csv each with different simulated missingness mechanisms. \n\nThe task is to predict the missing values in the test_A.csv and test_B.csv. The solution file should be formatted similarly to the sample_submission.csv. The Id column should include a string with RID_hash , VISCODE , col_name , test_set separated by underscores, for example `7cdf3031d8e8b148faf18742c72e6f21ffd63e9a344f729b9f33c7bd95e69355_90_CDRSB_test_B`. The Predicted column should include the predicted value.\n\n## Files\n\n* **dev_set.csv** - complete training set\n* **dev_1.csv** - training set with missing values, missingness mechanism #1\n* **dev_2.csv** - training set with missing values, missingness mechanism #2\n* **dev_3.csv** - training set with missing values, missingness mechanism #3\n\n* **test_A.csv** - test set with missing values, missingness mechanism A\n* **test_B.csv** - test set with missing values, missingness mechanism B\n\n* **sample_submission.csv** - a sample submission file in the correct format, predicting all missing values as the column means\n\n*Hint*: Participants will not be given information about missingness mechanisms A, B but are encouraged to develop algorithms that can characterize the missingness pattern and adapt the imputation approach accordingly.\n\n## Leaderboards\nOnly two submissions to the leaderboard are allowed per day. The public leaderboard contains 20% of all test_A samples and 20% of all test_B samples. The private leaderboard contains the remaining 80% of each dataset. Only the private leaderboard will be used to determine the challenge winner.\n\n## Columns\nEach row in the csv corresponds to a single subject visit. The data is longitudinal and all datasets (training set, public leaderboard set, and private leaderboard set) were split by subject. Ordinal or categorical variables will be treated as continuous during evaluation. Performance will be measured using range normalized mean absolute error (MAE).\n\nVariables that are always observed:\n* `RID_hash` - unique subject ID\n* `VISCODE ` - visit code, referring to the number of months\n\nVariables that may have missingness:\n* `AGE ` - age at that particular visit\n* `PTGENDER_num ` - sex {0: 'Male', 1: 'Female'}\n* `PTEDUCAT ` - education level\n* `DX_num ` - diagnosis {0: 'Cognitively Normal', 1: 'Mild Cognitive Impairment', 2: 'Dementia'}\n* `APOE4 ` - number of APOE e4 alleles\n* `CDRSB ` - cognitive score, Clinical Dementia Rating Sum of Boxes \n* `MMSE ` - cognitive score, Mini‐Mental State Examination\n* `ADAS13 ` - cognitive score, Alzheimer's Disease Assessment Scale-Cognitive Subscale\n* `Ventricles ` - ventricle volume \n* `Hippocampus ` - hippocampus volume \n* `WholeBrain ` - whole brain volume\n* `Entorhinal ` - entorhinal volume\n* `Fusiform ` - fusiform volume\n* `MidTemp ` - midtemp volume\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "**(!) Cash prizes of $10k, $5k, $2.5k available for 1st/2nd/3rd place submissions**\n\nScientists around the world are leveraging artificial intelligence to ask bigger questions and address previously intractable challenges in human biology and disease. Embracing AI tools, such as machine learning (ML), has become essential to accelerating drug discovery and development and meeting the needs of patients suffering from the most complex diseases. \n\nAs we continue to expand our [ML capabilities](https://www.gene.com/topics/advanced-computation-and-data-science), both internally and with external collaborators [external collaborators](https://www.gene.com/partners/what-we-are-looking-for/research-technologies), it is important to consider the existing hurdles that arise in preclinical and clinical research settings. One of the most common issues that scientists face is missing data. Gaps in data sets can bias the statistical analysis, creating a new host of problems down the road. \n\nIn line with our commitment to advancing AI/ML technologies, and in an effort to engage the broader ML community, we are launching a challenge to tackle the issue of missing data. ", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Task", "content": "To develop an imputation algorithm capable of handling different types of missingness in tabular datasets.\n\nFirst, the organizers will release the development data consisting of a complete dataset along with three missingness masks meant to simulate different patterns of missing data. Participants are encouraged to split this dataset as they see fit for training and validation.\n\nFinally, two held out test sets of equal size will be released, each with a different, unspecified missingness mechanism. The public leaderboard will include 20% of each test set and the private leaderboard will include the remaining 80% of each test set.\n\nWe expect the winning submissions will likely be an algorithm that can automatically characterize the pattern of missing data and adapt its imputation approach to the missingness type.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "The Competition Sponsor reserves the right to update the Competition Rules and Timeline throughout the duration of the Competition.\n\nENTRY IN THIS COMPETITION CONSTITUTES YOUR ACCEPTANCE OF THESE OFFICIAL RULES.\n\n###### *Accounts and Team Limits*
\nOnly one (1) account is allowed per participant, and there is a maximum of three (3) participants per Team. Team merging is allowed as long as the total participant count doesn't exceed the maximum. A Participant cannot create or join more than one (1) Team.\n\n###### *Submissions*
\nOnly two (2) Submissions to the public leaderboard are allowed within a day, and up to two (2) final Submissions will be accepted for judging. Submissions will be judged using the metric outlined in the *Overview>Scoring* section. For each team, a single final Submission with the best score on the Private Leaderboard will be considered for Prizes. The Submission must be received before the Competition Deadline as specified on the Kaggle Challenge page, and it is the responsibility of the Participant to determine the corresponding deadline in their jurisdiction. \n\nTo be Prize eligible, the Participant must deliver the code developed to generate the winning Submission to the Competition Sponsor for review. Submissions must be fully automatic, predictions submitted to the leaderboard cannot be manually labeled. The Competition Sponsor must be provided with the documentation necessary to replicate the winning Submission. The Competition Sponsor requires the code for final Submissions to be released under an Open Source License. Submissions must include a method description of at least 400 words and not more than 1,000 words (excluding references). Competition Sponsors reserve the right to verify participant eligibility and review code Submissions for compliance with the Competition Rules.\n\n###### *Prizes*
\nCompetition Title: Genentech 404 Challenge
\nCompetition Sponsor: Genentech Inc
\nTotal prizes available (Prize pool): $17,500 USD
\nFirst place: $10,000 USD
\nSecond place: $5,000 USD
\nThird place: $2,500 USD
\n\n\nPrizes will be awarded in the following order: first, second, and third place. In the unlikely event of a tie, the Competition Sponsor will equally divide the remaining Prizes until the Prize pool is depleted. \n\nWinners will be notified by email. Winners will be required to sign a Prize agreement and provide relevant tax documents within one (1) week of receiving an email notification. Documents include, but are not limited to, W9 for US Residents or a W-8BEN for foreign nationals. Prizes will be paid out approximately 30 days after the Competition Sponsor receives required documentation.\n\nIf a Team Submission results in a Prize, the Prize will be split equally between all Participants within the Team. All Participants within a Team must be eligible to receive a Prize and comply with all Competition Rules, failure to do so may result in the forfeiture of the Prize for the entire Team. \n\nFailure to comply with the above obligations by a winning Participant or Team of Participants will result in forfeiture of the Prize. In the case of a Prize forfeiture, the Competition Sponsor may select an alternate winner.\n\n###### *Eligibility*\nEntry into the competition constitutes acceptance of the Competition Rules.\n\nParticipants must be 18 years or older. Participant must have an account at Kaggle.com and an account at ida.loni.usc.edu. See Data section for more information about the terms related to the competition dataset. \n\nBy location: residents of Crimea, Donetsk People's Republic/Luhansk People's Republic, Cuba, Iran, Syria, and North Korea are not eligible to participate. Any individual or entity under US export controls or sanctions is not allowed to participate. Participants must comply with rules and regulations in their jurisdiction regarding participation in skills-based competitions.\n\nEmployees/affiliates of Genentech/Roche are allowed to participate in the Competition but are not eligible to win prizes.\n\n###### *Data*\nAll participants are required to sign and adhere to the Data Usage Agreement for the ADNI dataset on the LONI website (https://ida.loni.usc.edu/collaboration/access/appApply.jsp?project=ADNI). PARTICIPANTS SHOULD NOT DOWNLOAD THE COMPETITION DATA UNTIL THE ADNI DATA ACCESS REQUEST HAS BEEN APPROVED BY THE LONI WEBSITE ADMINISTRATORS. The Competition Data may not be shared or distributed in any form.\n\nParticipants are not allowed to re-identify the Competition Dataset with data from the LONI website to impute missing values in the test sets or to check the accuracy of their algorithms on the test sets (public or private leaderboard).
\n\nTraining on publicly available external data is allowed but must be reported in the method description. Training on ADNI data other than the Competition Dataset is not allowed. Moreover, training on synthetic data derived from ADNI data other than the Competition Dataset is not allowed.
\n\nNO PURCHASE NECESSARY TO ENTER.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Scoring & Prizes", "content": "Submissions will be evaluated using range normalized mean absolute error (MAE) to quantify the difference between the predicted values and the observed values. For the purposes of scoring, all variables will be treated as continuous variables and each missing instance will be given equal weight. \n\nThe final ranking will be determined by performance on the held out test sets only (private leaderboard). The private leaderboard will be revealed 7 days after submissions close and winners will be announced at the [UCSF Digital Health Hub Awards 2022](https://www.ucsfhealthhub.com/awards/2022/home) on November 14th. Participants are not required to attend the UCSF Digital Health Hub Awards.\n\nFirst place: $10,000 USD\nSecond place: $5,000 USD\nThird place: $2,500 USD\n\n• Final submissions must include a method description of at least 400 words and not more than 1,000 words (excluding references)\n• The Competition Sponsor requires the code for final Submissions to be released under an Open Source License ", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Data Access", "content": "**IMPORTANT! To participate in the 404 Challenge all participants MUST sign a Data Use Agreement from the LONI website (https://ida.loni.usc.edu/collaboration/access/appApply.jsp?project=ADNI).**
\nBy agreeing to the Competition Rules, participants confirm they have *requested and been granted access to* the ADNI database. Be aware that access may take between 3-5 business days.\n\nSee the Data tab for more information.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Organizing Team", "content": "The Genentech – 404 Challenge was organized by the Early Clinical Development Informatics group within Genentech Research and Early Development (gRED).\n\nThe organizing team would like to acknowledge Kyunghyun Cho PhD for his technical guidance on simulating data missingness. \n\nChallenge lead: Claudia Iriondo PhD", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }