The concept of using DNA to compute was first launched by L. M. Adleman in 1994. Since then, for nearly three decades, DNA computation has been developed from infancy to maturation by scientists around the world. However, elegant in design but clumsy in practical applications has been an arguable fact that prevents broader applications of DNA computation, especially in the field of biomedical science.
In this context, we try to make a step forward of using DNA computing towards biomedical applications. A computational classifier is first trained in silico using miRNA profiles from TCGA database, and followed by a computationally powerful but simple in situ molecular implementation scheme using DNA. However, although we have established a good machine learning model and reasonable molecular computation scheme, the low concentrations of miRNA in serum sample is not sufficient to trigger the signal reporting of molecular interactions.
There are many ways to rapidly amplify the nucleic acids, such as PCR and isothermal amplification. However, these amplification methods will typically generate ratio changes in the amplicon concentrations post the exponential amplification process. It should be noted that miRNA levels of individuals exhibit a great complexity in their variety and quantity, and alterations in either factor can lead to different diagnostic results. Furthermore, these amplicons are dsDNA, which is not suitable for our DNA molecular computation. Fortunately, we soon found a solution. A method called Linear-After-The-Exponential-PCR was introduced into the scheme for amplification of miRNA inputs in serum without significantly perturbing their original variety and quantity information
To build an effective classifier model, we selected the support vector machine (SVM) to train the miRNA-seq data from TCGA. As shown in Fig.1, the model including 4 miRNA inputs associated with appropriate weights were selected can realize NSCLC diagnosis by the formula ∑[c(miRNA n) * Wn] (n = 1, 2, 3, 4). To have simple, but accurate, diagnostic results, a systematic implementation of the SVM model with DNA computation scheme was designed, in which the mathematical weighted sum of healthy miRNA inputs is compared with that of cancer miRNA inputs, and only the larger is used to report the corresponding signals for diagnosis of either healthy or cancer.
Fig. 1. Work flow for cancer diagnosis with DNA computational platform.
In summary, we designed and tested a DNA computation-based cancer diagnostic method that can accurately differentiate NSCLC and healthy individuals from their serum samples. The idea we used to program this autonomous molecular system provides immediate possibilities in performing in situ diagnosis with advantages such as speed, simplicity, low-cost and less tendency for error. We believe that the ability of DNA to interact with naturally occurring biomolecules, together with such unique properties as programmability, nanometric size, and inexpensive equipment, can allow them being interfaced to other biological molecules, thereby solving problems in biological and biomedical research. We anticipate that the unique capabilities of DNA computation without human intervention on data analysis will justify the future effort in developing more powerful platforms. We envision that the power of DNA computing could inspire more clinical applications towards non-invasive and routine early cancer screening and classification, as well as monitoring cancer recurrence.
1. Zhang, C., Zhao, Y., Xu, X. et al. Cancer diagnosis with DNA molecular computation. Nat. Nanotechnol. (2020). https://doi.org/10.1038/s41565...
2. Adleman L. M. Molecular computation of solutions to combinatorial problems. Science, 266, 1021-1024 (1994).
3. Kevin M. Cherry, Lulu Qian. Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks. Nature 559, 370-376 (2018).
4. Lopez, R., Wang, R. & Seelig, G. A molecular multi-gene classifier for disease diagnostics. Nature Chemistry 10, 746–754 (2018).