Artificial intelligence screens out key proteins to identify benign and malignant thyroid nodules

Combining proteomic big data with artificial intelligence, we are committed to developing new methods of assisted clinical diagnosis based on proteomics and other molecular omics, in order to achieve accurate assisted diagnosis of major human diseases such as tumors.

It is understood that Xihu Omi's first project that is expected to achieve clinical transformation is the diagnosis of benign and malignant thyroid nodules based on protein markers.

Guo Tiannan, a distinguished researcher of theSchool of Life Sciences at West Lake University , said that proteomics is another breakthrough in human exploration of the mysteries of life activities following genomics and transcriptomics.

Capture subtle changes for protein weighing 

When people swallow, the organ that moves up and down with the swallowing action of the neck is the thyroid.

As early as 2010, the Endocrinology Branch of the Chinese Medical Association published the results of China's first epidemiological survey of thyroid diseases among community residents in ten cities, showing that the prevalence of thyroid nodules among residents was 18.

"According to this estimate, 1 out of every 5 adults today may have thyroid nodules.

"In the treatment of thyroid nodules, many people have to choose to remove the thyroid.

According to reports, a person's genes are unchanged in most cases from life to death, from health to disease.

Guo Tiannan said that all proteins, even if they can be seen with the naked eye, are difficult to accurately identify because of their complex structure.

"Weighing protein is the first step, and it is not enough to distinguish between benign and malignant thyroid nodules.

The research team built an algorithm model to calculate the 2,622 proteomics data to the power of about 2, and then score the 20 key proteins that were screened for benign and malignant.

Set algorithm score accuracy rate up to 89%

In the view of Professor Li Ziqing from the School of Engineering of West Lake University, identifying benign and malignant thyroid nodules from mass spectrometry data is like identifying "two" human faces with similar lengths from an image.

He led the research team to use machine learning methods to select 2622 meaningful candidate feature proteins from raw mass spectrometry data, and build a set of unique algorithm models suitable for proteomics data through neural network technology.

Li Ziqing introduced that using this set of models, the researchers scored the overall situation of these 20 proteins (scores between 0-1): when the comprehensive score is greater than or equal to 0.

"This technology can also be used to screen drugs for the treatment of tumors.

Guo Tiannan said that studies have shown that each tumor may have one or more of these abnormal proteins that cause disease, and may also change as the disease progresses.
If we can use proteomics + AI technology to find the abnormal protein at the current stage of the disease in real time, we can achieve "prescription of the right medicine".
It is believed that with the rapid development of molecular medicine, big data technology and artificial intelligence, proteomics will show increasing application potential in the field of precision medicine.