Liquid biopsy is moving into the era of artificial intelligence at high speed!

Firstly, tissue biopsy has a lagging nature. Second, tumors are heterogeneous, and for patients whose cancer cells have metastasized, only taking tumor tissue from a certain site does not reflect the overall situation of the patient. In addition, some patients are not suitable for tissue biopsy and some tumors are at risk of accelerated metastasis after interference from surgery, which is detrimental to patient care. in 2012, the U.S. Preventive Services Task Force (USPSTF) found that about one-third of biopsy patients develop complications. Among those biopsied for prostate cancer, they had a 20% to 30% risk of long-term effects such as urinary incontinence and erectile dysfunction.

First of all, liquid biopsy technology is cost-effective, and then it can dramatically reduce the difficulty of cancer diagnosis and care through non-invasive sampling, diagnosing cancer earlier and effectively prolonging patient survival.

We know that although the tumor lies quietly dormant in the body, the tumor cells will start metastatic migration by slipping into the vascular system and the vesicles secreted by the cancer cells and the free-floating DNA are released into the bloodstream. These cell fragments are a bunch of biomarkers that can signal cancer and predict its progression and response to treatment, and the use of liquid biopsies to detect and evaluate them is of great importance. Business consulting firm RNCOS estimates that the global liquid biopsy industry could exceed the $1 billion mark by 2020.

Several academic labs and biotech companies are already focusing on artificial intelligence, working to develop machine learning algorithms that can help decipher weak signals in the blood to identify cancer at an early stage and determine if it is responding to treatment.

Just like neurons in the brain, AI networks use neural networks, or thousands of connected nodes, to interpret data, and these networks can process large amounts of data and identify patterns that may be of interest to people. Not only that, but machine learning algorithms also improve themselves, fine-tuning the algorithm to improve its diagnostic sensitivity as more data is fed into the system.In May 2018, artificial intelligence (AI) genomics company Freenome, Inc. announced the first clinically validated study of a machine learning algorithm that screens DNA, RNA and protein sequences in the blood of high-risk patients to help diagnose colorectal cancer.

Just earlier this year, Landau's lab also announced a new machine learning approach for detecting cancer mutations in very low amounts of cell-free (cf) DNA, with the hope of using it to monitor cancer treatments. The algorithm compares whole genome sequences from tumor biopsy samples with mutation patterns in fragments of cfDNA extracted from blood. Scientists typically identify true tumor mutations by measuring the number of millions of repetitive DNA fragments containing a given mutation; the more fragments that agree, the more likely it is that the mutation is present in the tissue. But because there are so few fragments of cfDNA, Landau's software looks for complex mutation patterns throughout the sequence to estimate whether the fragments have been sequenced correctly. Non-small cell lung cancer mutations were detected in two patients using this algorithm with a sensitivity of 90%, significantly better than standard liquid biopsy techniques.