A neural network as a tool for diagnosis and treatment

Contents

• Using neural networks for genome interpretation

• Diagnosis and treatment of monogenic genetic disorders

• Huntington’s disease

• Hemophilia A

• Duchenne muscular dystrophy

• Polycystic kidney disease

• Identification of rare hereditary diseases using a person’s photo

• Development and future of this domain

Human DNA is an extremely important dataset which determines structure and functions of cells. This huge amount of data contains clues to treatment of both well-known and unknown diseases. Artificial intelligence (AI) helps identify hidden patterns material at a molecular level.

Using neural networks for genome interpretation

A gene is the basic physical and functional unit of heredity. Genes are made up of DNA¹. Genes bear information about physical and biological features. About 20,000 genes contain instructions for making proteins. Proteins have various functions and they are “building materials” for cells and tissues.

Genome is a complete set of genes in a cell of an organism. It contains all information necessary for development and functioning of a human body³. Scientists use neural networks to decipher unknown genomic variants.

The neural network makes genome sequencing (deciphering the DNA⁴) simpler:

• clinical genome interpretation is carried out quicker

• separate geneticvariants are annotated and recognized

• complex dependencies in the DNA sequence are identified.

The AI-based Fabric Gem platform makes genome interpretation quicker. The neural network finds links between inherited traits and clinical data. The aim is to find a gene which can be a cause of an unknown disease ⁵. The established diagnosis helps start treatment faster. This is especially important for babies with serious diseases in the first 24-48 hours of their lives⁶.

Diagnosis and treatment of monogenic genetic disorders

Monogenic disorders are caused by defects in a particular gene, for instance, due to mutations. As a result, a protein is coded and produced incorrectly, which manifests itself in different ways. It depends on which gene has a defect⁷.

Pieces of research where a neural network helped understand the nature of various monogenic disorders in order to select treatment is described below.

Huntington’s disease

Huntington’s disease causes degeneration of nerve cells in the brain. This results in problems with movement and thinking⁸.

The neural network makes genetic analysis quicker when the disease is identified. In a piece of research conducted in Spain scientists used the neural network to recognize the Huntington’s disease. How the neural network works⁹:

1. DNA methylation data are the input.

2. Based on chemical transformation of a DNA molecule, algorithms suggested whether there was no disease or there was the Huntington’sdisease.

3. The maximum sensitivity was 95%, and specificity was 80%.

Researchers believe that automated screening will help develop treatment for the disease⁹.

Hemophilia A

Hemophilia A is the genetic disorder characterized by a lack of blood clotting factor VIII (FVIII) or its defects. This can cause even small cuts to bleed for a long time.¹⁰.

Scientists from Brazil used the neural network to examine the protein structure of FVIII in patients with hemophilia A. It is assumed that differences in protein structure are related to severity of the disease. The algorithm helped separate neutral mutations from those which have a negative impact on FVIII function¹¹.

Duchenne muscular dystrophy

Duchenne muscular dystrophy (DMD) is a hereditary disease characterized by progressive loss of muscle. The disease is caused by a change in dystrophin. This is a protein which maintains muscle cell integrity¹².

Scientists from the University of Florida and the University of Colorado used the neural network to assess DMD progression¹³:

1. The input data was represented by MRI images. Using them, it was possible to see how the muscle tissue was gradually replaced by fat.

2. The neural networkrecognizes an area with considerable changes and assesses muscle texture.

3. The obtained result helps identify the subtype of the disease to adjust treatment.

The neural network precision was 91.7%. The method can be helpful to monitor a hereditary musculardisease¹³.

Researchers from China collected 2,536 MRI images for 432 cases of muscular diseases. They found genetic disorders (dystrophinopathies) in 148 cases. These disorders included DMD¹⁴.

The neural network selected an area of interest in the image and classified it as dystrophinopathy or another muscular disease. The digital model precision was 91%. The development can increase the importance of MRI in DMD diagnosis¹⁴.

Polycystic kidney disease

Polycystic kidney disease (PKD) is a genetic disorder. It is characterized by numerous fluid-filled cysts in kidneys. The volume of the affected organ increases, and there is deterioration in its function up to kidney failure¹⁵.

In one piece of research the neural network automatically measured the total kidney volume using MRI images. The result helps specialists to assess PKD severity, which is important for planning treatment¹⁶.

A radiologist usually highlights the anatomical region. It may take a lot of time to do this manually. The neural network helped do this 51% faster for 53 patients. If it was necessary, a specialist corrected automatic annotation results¹⁶.

Identification of rare hereditary diseases using a person’s photo

p>Scientists from the National Human Genome Research Institute Home (NHGRI) developed the neural network which helps identify six genetic disorders which manifest themselves on skin¹⁷:

1. The digital model was trained using panoramic images or separate parts of the body. Several images focused only on an affected area.

2. The model recognized how genetic disordersmanifested themselves on skin. That was done based on groups of pixels with unusual characteristics.

3. During tests precision was 81.4% in both cases.

Several genetic disorders are characterized by specific facial features. For instance, children with the Williams syndrome have a broad forehead, a small chin and a flat bridge of the nose ¹⁸.

In some cases changes in facial features are specific and can be easily recognized. in some other cases it can be difficult to establish a diagnosis. Neural network capabilities allow establishing the right diagnosis.

Scientists from China presented VGG-16, a facial recognition technology which identifies genetic syndromes using a photo. Children with 35 genetic disorders, including the Down syndrome and the Williams syndrome, took part in the research¹⁹.

The research was conducted as follows¹⁹:

1. The key areas were highlighted frontal face images.

2. The neural networkprocessed and recognized them, and identified the most probable syndrome.

3. The model results were compared with opinions of five pediatricians.

The neural network precision was 89%. The development will be helpful for screening of genetic syndromes in clinical practice¹⁹.

Specialists from the USA, Germany and South Africa developed a similar neural network called GestaltMatcher. It was trained and tested using images of 17,560 patients with 1,115 rare syndromes. The algorithm can identify even unknown diseases which were not included in the training set²⁰.

Face2Gene mobile app (USA) is another example. This is a tool for health professionals designed for a smartphone. Face2Gene is based on algorithms which involve deep learning and computer vision²¹. The neural network is trained to recognize 216 genetic disorders²².

Development and future of this domain

For successful use of neural networks and AI in genetics it is necessary to do the following²³:

• ensure confidentiality of information and secure data transfer for clinical trial and research

• create databases which will improve functioning of algorithms

• eliminate possible inaccuracy and bias in the data.

The future is related to development of technologies which will take into account complex causal links between genes and cells. This will enable prediction of changes in case of a particular gene variant. Scientists expect that AI will help develop treatment for rare and unknown diseases²⁴.

Sources

1. What is a gene? [Online resource]: MedlinePlus. URL:https://medlineplus.gov/genetics/understanding/basics/gene/.

2. A gene [Online resource]: National Human Genome Research Institute. URL:https://medlineplus.gov/genetics/understanding/basics/gene/.

3. Genome [Online resource]: National Human Genome Research Institute. URL: https://www.genome.gov/genetics-glossary/Genome.

4. Dias R, Torkamani A. Artificial intelligence in clinical and genomic diagnostics. Genome Med. 2019 Nov 19;11(1):70. doi: 10.1186/s13073-019-0689-8. PMID: 31744524; PMCID: PMC6865045.

5. Fabric Gem [Online resource]: Fabric Genomics. URL:https://fabricgenomics.com/fabric-gem/.

6. De La Vega FM, Chowdhury S, Moore B, Frise E, McCarthy J, Hernandez EJ, Wong T, James K, Guidugli L, Agrawal PB, Genetti CA, Brownstein CA, Beggs AH, Löscher BS, Franke A, Boone B, Levy SE, Õunap K, Pajusalu S, Huentelman M, Ramsey K, Naymik M, Narayanan V, Veeraraghavan N, Billings P, Reese MG, Yandell M, Kingsmore SF. Artificial intelligence enables comprehensive genome interpretation and nomination of candidate diagnoses for rare genetic diseases. Genome Med. 2021 Oct 14;13(1):153. doi: 10.1186/s13073-021-00965-0. PMID: 34645491; PMCID: PMC8515723.

7. Condò I. Rare Monogenic Diseases: Molecular Pathophysiology and Novel Therapies. Int J Mol Sci. 2022 Jun 10;23(12):6525. doi: 10.3390/ijms23126525. PMID: 35742964; PMCID: PMC9223693.

8. Huntington’s disease [Online resource]: Mayo Clinic. URL:https://www.mayoclinic.org/diseases-conditions/huntingtons-disease/symptoms-causes/syc-20356117.

9. Alfonso Perez G, Caballero Villarraso J. Neural Network Aided Detection of Huntington Disease. J Clin Med. 2022 Apr 10;11(8):2110. doi: 10.3390/jcm11082110. PMID: 35456203; PMCID: PMC9032851.

10. Hemophilia A [Online resource]: National Hemophilia Foundation. URL:https://www.hemophilia.org/bleeding-disorders-a-z/types/hemophilia-a.

11. Lopes, T.J.S., Rios, R., Nogueira, T. et al. Prediction of hemophilia A severity using a small-input machine-learning framework. npj Syst Biol Appl 7, 22 (2021). https://doi.org/10.1038/s41540-021-00183-9.

12. Duchenne muscular dystrophy [Online resource]: Muscular Dystrophy Association. URL:https://www.mda.org/disease/duchenne-muscular-dystrophy.

13. Cai J, Xing F, Batra A, Liu F, Walter GA, Vandenborne K, Yang L. Texture Analysis for Muscular Dystrophy Classification in MRI with Improved Class Activation Mapping. Pattern Recognit. 2019 Feb;86:368-375. doi: 10.1016/j.patcog.2018.08.012. Epub 2018 Sep 18. PMID: 31105339; PMCID: PMC6521874.

14. Yang M, Zheng Y, Xie Z, Wang Z, Xiao J, Zhang J, Yuan Y. A deep learning model for diagnosing dystrophinopathies on thigh muscle MRI images. BMC Neurol. 2021 Jan 11;21(1):13. doi: 10.1186/s12883-020-02036-0. PMID: 33430797; PMCID: PMC7798322.

15. Polycystic kidney disease [Online resource]: Mayo Clinic. URL:https://www.mayoclinic.org/diseases-conditions/huntingtons-disease/symptoms-causes/syc-20352820.

16. Goel A, Shih G, Riyahi S, Jeph S, Dev H, Hu R, Romano D, Teichman K, Blumenfeld JD, Barash I, Chicos I, Rennert H, Prince MR. Deployed Deep Learning Kidney Segmentation for Polycystic Kidney Disease MRI. Radiol Artif Intell. 2022 Feb 16;4(2):e210205. doi: 10.1148/ryai.210205. PMID: 35391774; PMCID: PMC8980881.

17. Duong D, Waikel RL, Hu P, Tekendo-Ngongang C, Solomon BD. Neural network classifiers for images of genetic conditions with cutaneous manifestations. HGG Adv. 2021 Aug 20;3(1):100053. doi: 10.1016/j.xhgg.2021.100053. PMID: 35047844; PMCID: PMC8756521.

18. Williams syndrome [Online resource]: MedlinePlus. URL:https://medlineplus.gov/genetics/condition/williams-syndrome/.

19. Hong D, Zheng YY, Xin Y, Sun L, Yang H, Lin MY, Liu C, Li BN, Zhang ZW, Zhuang J, Qian MY, Wang SS. Genetic syndromes screening by facial recognition technology: VGG-16 screening model construction and evaluation. Orphanet J Rare Dis. 2021 Aug 3;16(1):344. doi: 10.1186/s13023-021-01979-y. PMID: 34344442; PMCID: PMC8336249.

20. GestaltMatcher [Online resource]: GestaltMatcher. URL:https://www.gestaltmatcher.org.

21. Face2Gene — How it works [Online resource]: Face2Gene. URL:https://www.face2gene.com/technology-facial-recognition-feature-detection-phenotype-analysis/.

22. Qiang J, Wu D, Du H, Zhu H, Chen S, Pan H. Review on Facial-Recognition-Based Applications in Disease Diagnosis. Bioengineering (Basel). 2022 Jun 23;9(7):273. doi: 10.3390/bioengineering9070273. PMID: 35877324; PMCID: PMC9311612.

23. Alrefaei AF, Hawsawi YM, Almaleki D, Alafif T, Alzahrani FA, Bakhrebah MA. Genetic data sharing and artificial int ap Appleelligence in the era of personalized medicine based on a cross-sectional analysis of the Saudi human genome program. Sci Rep. 2022 Jan 26;12(1):1405. doi: 10.1038/s41598-022-05296-7. PMID: 35082362; PMCID: PMC8791994.

24. König H, Frank D, Baumann M, Heil R. AI models and the future of genomic research and medicine: True sons of knowledge?: Artificial intelligence needs to be integrated with causal conceptions in biomedicine to harness its societal benefits for the field. Bioessays. 2021 Oct;43(10):e2100025. doi: 10.1002/bies.202100025. Epub 2021 Aug 11. PMID: 34382215.