Mars Spectrometry 2: Gas Chromatography

Help NASA scientists identify the chemical composition of rock and soil samples for Mars planetary science. #science

$30,000 in prizes
oct 2022
537 joined

Mass spectrometers are now, and will continue to be, a key instrument for missions searching for life and habitability on other planets.

— Victoria Da Poian, NASA Goddard Space Flight Center Data Scientist & Engineer

Why

Did Mars ever have livable environmental conditions? NASA missions like the Curiosity and Perseverance rovers carry a rich array of instruments that can collect data — such as the chemical makeup of rock and soil samples — to help build evidence around this question.

However, these instruments cannot currently analyze samples automatically. This capability could help missions to guide science operations, reduce reliance on "ground-in-the-loop" analysis, and prioritize transmission over increasingly long distances.

The Solution

The goal of this research challenge was to to build a model to automatically analyze data collected using a method of chemical analysis—gas chromatography–mass spectrometry (GCMS)—that is performed by the Curiosity rover's SAM instrument suite on soil and rock samples. Competitors were tasked with detecting certain families of chemical compounds that are of scientific interest in analyzing conditions for past habitability.

This challenge built on the previous DrivenData competition Mars Spectrometry: Detect Evidence for Past Habitability, where participants developed approaches for automated analysis of data collected using evolved gas analysis (EGA), another type of chemical analysis performed by the SAM instrument suite.

The Results

The winning solution achieved a leaderboard-topping log loss score (lower is better) of 0.1443 and average precision (higher is better) of 0.81. Below is an example of this model in action, analyzing the left sample and generating the predictions on the right:

Example sample.
aromatic 0.008592
hydrocarbon 0.975277
carboxylic_acid 0.006051
nitrogen_bearing_compound 0.004336
chlorine_bearing_compound 0.000584
sulfur_bearing_compound 0.001165
alcohol 0.001052
other_oxygen_bearing_compound 0.002319
mineral 0.017754
Note: compounds in blue were present in the sample.

The top competitors also submitted write-ups that explained and illustrated their approaches, including visualizations of their architectures and intermediate steps. All of these write-ups can be found alongside the winners' models in the publicly accessible winners' repository for this competition.


RESULTS ANNOUNCEMENT + MEET THE WINNERS

WINNING MODELS ON GITHUB

GCMS OPEN DATASET

EGA OPEN DATASET