Team:Patras/Proof Of Concept
Proof Of Concept
- Wet Lab
Our first goal was to prove that the BentoLab can provide reliable results in genomic analysis. Therefore, all samples were genotyped using the KASP assay, using the Hydrocycler-4 (LGC Genomics) coupled to FLUOstar Omega SNP (BMG LABTECH), as the gold standard genotyping method, which was standardized against Sanger sequencing methodology.
Το further validate our in vitro findings, 81 samples were genotyped in a conventional laboratory device, and a comparison between them and the findings from BentoLab was conducted, using again as gold standard the results produced from the KASP assay.
We performed genotyping analysis for 81 samples using both methods, the conventional one, and the BentoLab based on the same principle. Bellow, we present the results for an indicative number of these samples. More precisely, Table 1 represents the genotypes of four samples derived from the gold standard method and the performance of the other two methods (the conventional benchtop laboratory equipment and the BentoLab), which produced highly accurate results. Moreover, as shown in Figure 1, we also present the results from AS-PCR as performed for four samples and by following the protocols for the gold standard method, the conventional benchtop laboratory method, and the BentoLab method.
Table 1: Representation of the genotypes of 4 different DNA samples (A, B, C, D) for rs4149056, according to the gold standard method, the conventional benchtop laboratory equipment, and the BentoLab. Anc. allele stands for the ancestral allele, Alt. allele for the alternative one, and GT indicates the genotype.
| Sample ID | Anc. allele | Alt. allele | GT in gold standard method | GT in Conventional laboratory equipment | GT in BentoLab |
|---|---|---|---|---|---|
| 'A' | T | C | T|T | T|T | T|T |
| 'B' | T | C | T|C | T|C | T|C |
| 'C' | T | C | C|C | C|C | C|C |
| 'D' | T | C | T|T | T|T | T|T |
Figure 1: Representation of the rs4149056 genotype for four different DNA samples (A, B, C, and D), as also described in Table 1, by using (A) the static laboratory equipment and (B) the BentoLab method. For each of the DNA samples, two wells match: for the wild type (“w,” referring to T allele) and for the mutant allele (“m,” referring to C allele). Ladder 100 bp was used, and the expected molecular weight per PCR product is 535 bp. DNA, deoxyribonucleic acid.
To make the genotype results countable, we established a system based on genotype and score predictions, both derived for each sample and each method assigned by 11 independent qualified evaluators. The evaluators were asked to identify the genotype and rate each sample (N = 81 samples) from a scale of 1 to 4, depending on the occurrence of the respective band within all the agarose gels produced. Specifically, score 1 is given to the samples where the genotype is difficult to be detected or incorrect compared to the gold standard method and needs to be re-examined. Accordingly, score 2 is given to samples with a dubious genotype, score 3 to samples with a correct genotype, which could be accompanied by a faint non-specific product, i.e., a faint band of DNA on the well of the mutant allele in a sample T/T and a faint band on the well of the wild type allele in a sample C/C. Finally, score 4 is given to samples where the correct genotype is easily detected, and the error probability is reduced.
Moreover, each sample’s median value was calculated to provide a representative value of all 11 independent observations. For the comparison between the two samples, we implemented a Fisher’s test by creating a 2x2 table in GraphPad software and calculating the p-value. Regarding the scoring assessment of the 81 samples, the percentages of concordance between the three methods were also estimated. When it comes to the BentoLab method, 94% (76 out of 81samples) of concordance against the gold standard method was observed for samples with scores 3 and 4, which were the samples with a genotype that can be used for further statistical assessment. Regarding the conventional benchtop laboratory method, 77 out of 81 samples (95%) were correctly identified according to the gold standard method (i.e., samples with scores 3 and 4). Fisher’s exact test was conducted to assess the significance of differences between the two genotyping methods. As described above, scores 1 and 2 corresponded to the FALSE values, whereas scores 3 and 4 corresponded to the TRUE values. We hypothesized that the genotyping ability of BentoLab and the conventional benchtop method is no different, as expected since both were thermal cyclers based on the Peltier effect. The p-value was higher than 0.05, so our null hypothesis is not being rejected, and the association between the two examined methods is not statistically significant. Therefore, our proposed BentoLab-based method can successfully produce results comparable with a benchtop laboratory method and the gold standard, while future studies in larger samples are warranted. The comparison between the BentoLab and the benchtop laboratory equipment is represented below in Figure 2.
Figure 2: Comparison between the BentoLab method and the conventional method according to the median scores for the 81 samples, as derived from the assigned quality scores.
To further estimate the BentoLab method’s validity and the conventional genotyping approach, ten randomly selected samples were tested seven times using either the BentoLab method or the conventional benchtop laboratory method. As shown in Figure 3, none of the ten samples deviated from the standard flat joint line.
Figure 3: Repeats for the BentoLab method after running AS-PCR for ten different samples seven times each. Each one of the 7 times run per sample was a complete AS-PCR assay conducted. The spots on the standard flat joint line represent the samples with the correct genotype. The dots that deviate from the standard flat joint line represent samples that gave wrong results when repeating the experiment. AS-PCR, allele-specific polymerase chain reaction.
This study provides evidence that the BentoLab is as accurate as any conventional benchtop laboratory method while also allowing the identification of the correct genotype in a high percentage of samples and repeats. Consequently, the BentoLab method is proposed as a valid and reliable method for genotyping in clinical practice for further consideration and evaluation in health settings.
The DOI number is also provided for a quicker search: DOI: 10.1089/omi.2020.0168.
*These two authors contributed equally to this work.
1. Laboratory of Pharmacogenomics and Individualized Therapy, Department of Pharmacy, School of Health Sciences, University of Patras, Patras, Greece.
2.Department of Pathology, College of Medicine and Health Sciences, United Arab Emirates University, Al-Ain, United Arab Emirates.
3.Zayed Center
- Dry Lab
Even though our project’s main goal is to prove that health-related problems can be solved using AI techniques, our team tried to create Deep Learning models to generalize their knowledge and perform accurately in images that they are not trained in. In addition, we created new images, using a photo editing program, photoshop, to create more training data to improve our model’s ability to detect the OoI correctly in new images.
Initially, we had a small set of images (24) from the electrophoresis gel result, and we used them as a training set to validate the system. When we split the dataset into training and testing set by using 19 and 5 images, respectively, we find out that the network can’t make predictions in images that are not trained in, even though they look similar. This happened because the instances’ distribution was unbalanced, and the pictures were taken from different height with varying conditions of light. When we created our dataset with 5.814 images, we tested our model accuracy in 18 test images. These images are from the original 5 images of the 24 in total images we possessed. We also test our system results by creating new images, using these 5 ones, at different crops and zoom levels. Still, our model was able to correctly predict the position and class of the DNA fragment bands in the picture in most cases.
Finally, our BentoLab-AI system was tested by three medical students who followed the step-by-step guidance to conduct the experiment and upload the photo from the electrophoresis gel to get the dose recommendations. We tested the image using different crops and zoom-level pictures, and we had 100% accurate results most of the time.
Sponsors
About
16 students from Patras blending Pharmacogenomics with Artificial Intelligence to redefine medicine
Follow Us
© iGEM Patras - All Rights Reserved