top of page

AI: the next step in diagnosis and treatment of genetic diseases

AI can process data sets and identify patterns and biomarkers

With the development of more intricate Artificial Intelligence (AI) software, which has rapidly grown from the chaotic chatbots to the more well-formed ChatGPT, it is easy to think we are seeing the rise of powerful artificial intelligence that could potentially replace us all. However, there is one problem. Originality does not exist for AI, at least not complete originality.

At its most basic, an AI program is trained on a set of data, whether this be an entire search engine’s worth of data, as is the case for ChatGPT, or a few images and phrases gathered from the internet. Therefore, an AI does not know any more than what it can quote or infer from the provided data, which means that a piece of art, a picture of a family, or any short story AI is asked to produce is often a replica of techniques or a chaotic and terrifying mess of images it has been given to use. However, here also lies its strength. AI can take in thousands of images and data sets and notice minor changes and differences the average person could not. It is, therefore, not AI’s ability to create the unique, but instead its ability to recognise the mundane that we can utilise, even in diagnosing and treating genetic disorders.


By analysing PET, MRI, fMRI and genetic data, AI can process enormous data sets and identify subtle patterns and biomarkers that often elude human observations, enabling earlier and more precise diagnosis. When looking at examples of the application of AI in the diagnosis of genetic disorders, a good reference is the so-far successful use of AI in diagnosing Huntington’s disease.

Huntington’s disease diagnosis using AI

Huntington’s disease symptoms present as patients experience involuntary movements and a decline in decision-making processes. Huntington's disease is a genetic disorder, meaning it is caused by a faulty gene, in this case, a fault in the Huntingtin gene (Htt).

The Huntington’s disease mutation in Htt results from CAG trinucleotide repeats, a highly polymorphic expansion of Htt consisting of the CAG (cytosine, adenine, guanine) nucleotides (DNA building blocks). Whilst CAG repeats are common and often normal and unharmful, individuals with Huntington’s disease possess an abnormally high number of these CAG repeats (more than 36). When an individual has an abnormally high number of CAG repeats, their Htt proteins do not fold into their proper shape, causing them to bond with other proteins and become toxic to a cell, which ultimately causes cell death in crucial medium spiny neurons (MSN) in the basal ganglia. Basal ganglia are brain structures responsible for the fine-tuning of our motor processes, which they do by essentially allowing neurons to respond in a preferred direction (a target muscle) rather than a null direction using MSNs.

So, it is clear how Huntington's disease symptoms occur; mutant Htt leads to cell death in MSNs, leading to the basal ganglia’s inability to control movement, which causes characteristic involuntary behaviours, among other symptoms.

Because we identified these changes in Htt and loss of MSN in the basal ganglia, PET, MRI, and fMRI scans are often used in the diagnosis of Huntington’s disease, in addition to genetic and mobility tests. By collecting and extracting clinical and genetic data, certain AI algorithms can analyse the broad range of Huntington’s disease clinical manifestations, identify differences, including even minute changes in the basal ganglia that a doctor may not have, and make an earlier diagnosis. One branch of AI that has proved effective is machine learning.

Machine learning models in diagnosis

Machine learning uses data and algorithms to imitate the way humans learn. For Huntington's disease diagnosis, this involves the identification of biomarkers and patterns in medical images, gene studies and mobility tests, and detecting subtle changes between data sets, distinguishing Huntington’s disease patients from healthy controls. While machine learning in Huntington’s disease diagnosis comes in many forms, the decision tree model, where the AI uses a decision tree as illustrated in the Project Gallery, has proven very effective.

A decision tree model looks at decisions and their possible consequences and breaks them into subsets branching downward, going from decision to effect. Recent research using AI in Huntington’s disease diagnosis has utilised this model to analyse gait dynamics data. This data looks at variation in stride length, how unsteady a person is while walking, and the degree to which one stride interval (the time between strides) differs from any previous and any subsequent strides. For an individual, it is widely accepted that if they have abnormal variations in stride (their walking speed is reduced, their stance is widened), then they are exhibiting symptoms of Huntington’s disease. Therefore, by using this gait data, and having the machine learning model come up with a mean value for stride variation for trial patients, it will be able to discern which patients have stride variation associated with Huntington’s disease (a higher variation in stride) and those that do not. Researchers found that using this method of diagnosis, they were able to accurately identify which gaits belonged to Huntington's disease patients, with an accuracy of up to 100%. Furthermore, researchers also found decision tree models useful when identifying whether a gene links with Huntington’s disease when comparing patients' genetic information with prefrontal cortex samples, with this method’s accuracy being 90.79%.

With these results and even more models showing incredible promise, AI is already proving itself useful when it comes to identifying and diagnosing sufferers of genetic disorders, such as those with Huntington’s disease. But this leads us to ask, can AI even help in the treatment of those suffering from genetic disorders?

Treatment- current studies in cystic fibrosis

While AI models can be applied diagnostically for disorders such as Huntington's disease, they may also be relied upon in disease treatment. The use of AI in tailored treatment is the focus of current research, with one even looking at improving the lives of those suffering from cystic fibrosis.

Around 10,800 people are recorded as having cystic fibrosis in the UK, and this debilitating disorder results in a buildup of thick mucus, leading to persistent infections and other organ complications. The most common cause of cystic fibrosis is a mutation in the gene coding for the protein CFTR, resulting from a deletion in its coding gene, causing improper folding in the protein CFTR, as we saw in Huntington’s disease. This misfolding leads to its retention in the wrong place in a cell, so it can no longer maintain a balance of salt and water on body surfaces. Because of the complex symptoms arising from this imbalance, this disease is very difficult to manage, but there is hope, and hope comes as SmartCare.

SmartCare involved home monitoring and followed 150 people with cystic fibrosis for six months, having them monitor their lung function, pulse, oxygen saturation and general wellness and upload recorded data to an app. Subsequently, researchers at the University of Cambridge used machine learning to create a predictive algorithm that used this lung, pulse, and oxygen saturation data, identifying patterns that were associated with a decline in a patient's condition, and then predicted this decline much faster than the patient of their doctor could. On average, this model could predict a decline in patient condition 11 days earlier than when the patient would typically start antibiotics, allowing health providers to respond quicker and patients to feel less restricted by their health. This project was, in fact, so successful that the US CF Foundation is now supporting a clinical implementation study, called Breath, which began in 2019 and continues to this day.

Although there is a long way to go, using AI, the future can seem brighter. In Huntington’s disease and cystic fibrosis, we can see its effectiveness in both disease diagnosis and treatment. With the usage of AI predicted to increase in the future, there is a great outlook for patients and an opportunity for greater quality of care. This ultimately could ease patient suffering and prevent patient deaths. All this positive research tells us AI is our friend (although science fiction would often persuade us otherwise), and it will guide us through the tricky diagnosis and treatment of our most challenging diseases, even those engrained in our DNA.

Written by Faye Boswell

Related article: AI in drug discovery

Project Gallery

bottom of page