Open fpfcap opened 2 months ago
2024.05.28 Last Week: finish the review. Summarize totally 35 articles mentioned in the article and figure out the categories of the data mentioned on the review. According to the article mentioned in the last meeting----- DISMIR: Deep learning-based noninvasive cancer detection by integrating DNA sequence and methylation information of individual cell-free DNA reads-----there are 5 similar articles found. Still working with those 6 articles right now
Finish the draft of the documentation in Word, still learning the grammar of LaTeX, the final version will be revised on Latex. And approximately finishing the paper "DISMIR"
i've tried to use zotero and everything successful until i try to cite the article. There's a mistake i show below.
The citation seems wrong in somewhere but i don't know. Does this common? @Milokita
Plz use the template provided, you may failed to include the file in your tex
黄圈所示的输出门构造中t-1时刻的h和t时刻的输入x构成t时刻的h合理,但此中t时刻的C又起着什么作用呢?为什么他会参与t时刻的h的构成?
@Milokita
I retry the loading and find it's still the same. The figure below is the screenshot from the latex showing successfully refreshing but the result in the right still remain the same. The reference could be download and opened on zotero successfully.(white button next to the "refresh")
Last Week 1.Finishing the paper DISMIR, but still some questions about the data analysis left 2.Working with another paper CpG Transformer (report tomorrow) 3.Finishing collecting the DISMIR code, including the training used data. Next Week Plan 1.Trying to run the DISMIR code successfully 2.Compare the data analysis steps between two paper 3.Mixing the previous autoencoder code with DISMIR code after finishing the running steps
First.
it's from page 5, describing how to visualize the kernels. But actually i'm so curious about the process tracing back to the specific location on original sequence. For instance, if i give a read and the final d-score show significance, so if we wanna find out where the read is from, do we need to trace the original read first, and then find its location in the reference genome?
Second
It's about kernel, since the kernel would change with running of convolution layer. Does there any princple exit to decide how many kernel we should set in order to make the network do a suitable job and not overfitting.
@Milokita
figure out the gradient problem as well as the multiple labels problem with their solution
关于DISMIR的模型结合到自己的project里面需要进一步确定效果图如何@fpfcap
Last Week Work: finish finding two omics data: methylation, CNV . each of them contain the PRAD and BRCA data. They are all from TCGA. Next Week Plan: 1.the whole data still lack cnv control group data, i need to solve. 2.All the collected data is raw and shows original position, like cg000065, i need to associate it with the specific gene to firstly optimized. 3.To find if the mutation could be the features to used in the deeplearning, maybe use one-hot code(mentioned in the bsc6022).