HT24
Aim: The core of this project addresses the question, if an AI-trained gas sensor can be employed to identify the presence of a specific disease – like IBD – in the volatile compounds emitted from human fecal matter.
Background: Patients with inflammatory bowel disease (IBD) suffer from e.g., tummy pain, diarrhea, fecal bleedings, and weight loss. IBD usually is confirmed by biopsies on colonoscopy which is inconvenient for the patient, expensive and only available in hospitals/medical facilities by trained medical personnel. A medical device developed with the trained sensor facilitates detection of fecal bleedings as e.g., in IBD at home/without access to medical facilities, making it more convenient for the patient and achieving more equity in diagnostics for all people worldwide.
Methods:
• Apply a commercially available gas sensor: pre-tests will be run to test and optimize the sensor parameters as well as the sample preparation to create a library of different measurement conditions and sample preparations
• Machine learning, use part of the created data for algorithm training and part of it for validation. Apply different algorithms and compare them to the results from the Bosch algorithm
My task: The measuring device including the sensor will be provided for the student. The majority of the tasks will be experimental: Create a library of fingerprint smells of samples containing of stool alone or stool spiked with a compound of interest. Compounds to be tested are blood and 1-2 IBD biomarkers to see whether they can be detected and at which concentrations. Parameters to investigate will be e.g., concentrations of blood and stool, measurement temperature and time, and sample preparation (freezing, mixing, dilution with water/buffer). As a control, blood and calprotectin will be detected by the commercially available tests occult blood test Hemo-Fec and occult calprotectin test CalproSmart self-test (results partially provided from collaborators at Akademiska). All experimental results will be fed into machine learning using either the algorithm provided by the commercially available sensor or algorithms established in the group with support from Assoc. Prof. Per Larsson. If time permits, the trained and optimized algorithm can be tested on stool samples from IBD patients.
Expected results: Find optimized conditions for sensor parameters and sample preparation to increase accuracy of the AI to distinguish between healthy and diseased stool samples. The proof-of-principle includes also spiking the stool samples with biomarkers to see whether the sensor can detect a combination of different markers to get a unique fingerprint smell specific for the disease IBD. If the training of the algorithm is completed, testing it in real patients’ samples will either verify or falsify the hypothesis that an AI-trained gas sensor can be employed to identify the presence of a specific disease – like IBD – in the volatile compounds emitted from human fecal matter.
Läkemedelsutveckling
Läkemedelsformulering och Molekylär galenisk farmaci
Laborativ studie
Uppsala Universitet
Uppsala
Hannah Pohlit
Hannah.pohlit@farmaci.uu.se
Institutionen för farmaci
Masterprogram i läkemedelsutveckling
Degree project in Drug Discovery and Development 45 c -3FK044
45hp
1