#METABOLOMICS WORKBENCH Stopka28_20211206_115314 DATATRACK_ID:2969 STUDY_ID:ST002045 ANALYSIS_ID:AN003329 PROJECT_ID:PR001292 VERSION 1 CREATED_ON January 3, 2022, 5:26 pm #PROJECT PR:PROJECT_TITLE massNet: integrated processing and classification of spatially resolved mass PR:PROJECT_TITLE spectrometry data using deep learning for rapid tumor delineation PR:PROJECT_TYPE integrated processing and classification of MSI data analysis PR:PROJECT_SUMMARY The attached MSI datasets of GBM and prostate cancer tissues were analyzed in PR:PROJECT_SUMMARY the manuscript by Abdelmoula et al. (bioRxiv 2021.05.06.442938). The below is PR:PROJECT_SUMMARY taken from the abstract: "Motivation: Mass spectrometry imaging (MSI) provides PR:PROJECT_SUMMARY rich biochemical information in a label-free manner and therefore holds promise PR:PROJECT_SUMMARY to substantially impact current practice in disease diagnosis. However, the PR:PROJECT_SUMMARY complex nature of MSI data poses computational challenges in its analysis. The PR:PROJECT_SUMMARY complexity of the data arises from its large size, high dimensionality, and PR:PROJECT_SUMMARY spectral non-linearity. Preprocessing, including peak picking, has been used to PR:PROJECT_SUMMARY reduce raw data complexity, however peak picking is sensitive to parameter PR:PROJECT_SUMMARY selection that, perhaps prematurely, shapes the downstream analysis for tissue PR:PROJECT_SUMMARY classification and ensuing biological interpretation. Results: We propose a deep PR:PROJECT_SUMMARY learning model, massNet, that provides the desired qualities of scalability, PR:PROJECT_SUMMARY nonlinearity, and speed in MSI data analysis. This deep learning model was used, PR:PROJECT_SUMMARY without prior preprocessing and peak picking, to classify MSI data from a mouse PR:PROJECT_SUMMARY brain harboring a patient-derived tumor. The massNet architecture established PR:PROJECT_SUMMARY automatically learning of predictive features, and automated methods were PR:PROJECT_SUMMARY incorporated to identify peaks with potential for tumor delineation. The PR:PROJECT_SUMMARY model’s performance was assessed using cross-validation, and the results PR:PROJECT_SUMMARY demonstrate higher accuracy and a 174-fold gain in speed compared to the PR:PROJECT_SUMMARY established classical machine learning method, support vector machine." PR:INSTITUTE Brigham and Women’s Hospital PR:DEPARTMENT Department of Neurosurgery PR:LABORATORY Nathalie Y.R. Agar PR:LAST_NAME Abdelmoula PR:FIRST_NAME Walid PR:ADDRESS 60 Fenwood RD, Boston, MA PR:EMAIL wahassan@bwh.harvard.edu PR:PHONE 8572149765 #STUDY ST:STUDY_TITLE massNet: integrated processing and classification of spatially resolved mass ST:STUDY_TITLE spectrometry data using deep learning for rapid tumor delineation ST:STUDY_SUMMARY The patient-derived xenograft (PDX) mouse brain tumor model of glioblastoma ST:STUDY_SUMMARY (GBM) samples were analyzed by 2D MALDI FT ICR MSI. ST:INSTITUTE Brigham and Women’s Hospital ST:DEPARTMENT Department of Neurosurgery ST:LABORATORY Nathalie Y.R. Agar ST:LAST_NAME Abdelmoula ST:FIRST_NAME Walid ST:ADDRESS 60 Fenwood RD, Boston, MA ST:EMAIL wahassan@bwh.harvard.edu ST:PHONE 8572149765 #SUBJECT SU:SUBJECT_TYPE Mammal SU:SUBJECT_SPECIES Mus musculus SU:TAXONOMY_ID 10090 #FACTORS #SUBJECT_SAMPLE_FACTORS: SUBJECT(optional)[tab]SAMPLE[tab]FACTORS(NAME:VALUE pairs separated by |)[tab]Raw file names and additional sample data SUBJECT_SAMPLE_FACTORS - 1PDX GBM - mouse brain tumor section sample_id:Dataset1 RAW_FILE_NAME=GBM12_1 SUBJECT_SAMPLE_FACTORS - 2PDX GBM - mouse brain tumor section sample_id:Dataset2 RAW_FILE_NAME=GBM12_2 SUBJECT_SAMPLE_FACTORS - 3PDX GBM - mouse brain tumor section sample_id:Dataset3 RAW_FILE_NAME=GBM22_1 SUBJECT_SAMPLE_FACTORS - 4PDX GBM - mouse brain tumor section sample_id:Dataset4 RAW_FILE_NAME=GBM22_2 SUBJECT_SAMPLE_FACTORS - 5PDX GBM - mouse brain tumor section sample_id:Dataset5 RAW_FILE_NAME=GBM39_1 SUBJECT_SAMPLE_FACTORS - 6PDX GBM - mouse brain tumor section sample_id:Dataset6 RAW_FILE_NAME=GBM39_2 SUBJECT_SAMPLE_FACTORS - 7PDX GBM - mouse brain tumor section sample_id:Dataset7 RAW_FILE_NAME=GBM108_negative SUBJECT_SAMPLE_FACTORS - 8PDX GBM - mouse brain tumor section sample_id:Dataset8 RAW_FILE_NAME=GBM108_positive #COLLECTION CO:COLLECTION_SUMMARY As stated in the massNetpaper: "Briefly, 8 GBM tissue sections of 12 μm CO:COLLECTION_SUMMARY thickness were prepared and analyzed using a 9.4 Tesla SolariX mass spectrometer CO:COLLECTION_SUMMARY (Bruker Daltonics, Billerica, MA) in the positive ion mode with spatial CO:COLLECTION_SUMMARY resolution of 100 μm. The MSI data was exported from SCiLS lab 2020a (Bruker, CO:COLLECTION_SUMMARY Bremen, Germany) in the standardized format imzML (Race et al., 2012) and CO:COLLECTION_SUMMARY converted to the HDF5 format (Folk et al., 2011) for deep learning analysis." CO:SAMPLE_TYPE PDX GBM - mouse brain tumor section #TREATMENT TR:TREATMENT_SUMMARY N/A #SAMPLEPREP SP:SAMPLEPREP_SUMMARY As stated in the massNet paper "Briefly, 8 GBM tissue sections of 12 μm SP:SAMPLEPREP_SUMMARY thickness were prepared and analyzed using a 9.4 Tesla SolariX mass spectrometer SP:SAMPLEPREP_SUMMARY (Bruker Daltonics, Billerica, MA) in the positive ion mode with spatial SP:SAMPLEPREP_SUMMARY resolution of 100 μm. The MSI data was exported from SCiLS lab 2020a (Bruker, SP:SAMPLEPREP_SUMMARY Bremen, Germany) in the standardized format imzML (Race et al., 2012) and SP:SAMPLEPREP_SUMMARY converted to the HDF5 format (Folk et al., 2011) for deep learning analysis." #CHROMATOGRAPHY CH:CHROMATOGRAPHY_TYPE None (Direct infusion) CH:INSTRUMENT_NAME none CH:COLUMN_NAME none #ANALYSIS AN:ANALYSIS_TYPE MS #MS MS:INSTRUMENT_NAME Bruker Solarix FT-ICR-MS MS:INSTRUMENT_TYPE FT-ICR MS:MS_TYPE MALDI MS:ION_MODE POSITIVE MS:MS_COMMENTS Bruker software MS:MS_RESULTS_FILE ST002045_AN003329_Results.txt UNITS:Da Has m/z:Yes Has RT:No RT units:No RT data #END