Hello everyone,
Can I use _truth position_ during training as same as labels for cells in 1297 cells data? Because if you follow 3 threads: [Actual position labels](https://www.synapse.org/#!Synapse:syn15665609/discussion/threadId=3480), [Approach](https://www.synapse.org/#!Synapse:syn15665609/discussion/threadId=3476), [DistMap](https://www.synapse.org/#!Synapse:syn15665609/discussion/threadId=3477), it seems like I'm allowed to use that. Now, I'm still training unsupervised only based on RNA expression for 1297 cells and 60/40/20 in-situ genes information but can use it I can improve my algorithms.
Thanks all.
Created by Mr Robot mrrobot2018 Thanks for your support! I got it! Sorry your example is unclear, let me phrase it again.
1)You can use the "true" position from the 1297 cells as calculated by the maximum MCC using the 84 ins situs.
2) You can use the in situ position information for 20/40/60 genes from the 84 in situs
3) You cannot select the 20/40/60 genes in situs from the 84 genes by using all 84 genes and thoroughly due a feature selection based on their positional information relative to the "true" position. I've misunderstood by your answer.
For examples:
In-situ matrix:
```
Genes: A B C D E
Position 1: 1 0 0 0 0
Position 2: 1 1 1 1 1
Position 3: 1 0 1 1 1
```
Assume, If I choose genes **[A, D]** then can I use matrix
```
([1 0] [1 1] [1 1])
```
for it? Because, if the predicted model only based on _truth position_ (maximum 1297 positions ) then we don't have the information for the position other. How can I get information about the other position not in _truth_? You can get the in situ matrix, but you have to select the 20/40/60 genes from the 84 without using it.
Furtheremore there can be many ways to predict cell positions without using any in situ data (besides for getting the positions) , i.e selecting features from RNA-seq to predict positions.
P Hi @jeriscience.
If I can't get sub-in situ matrix for 20/40/60 genes we chose, then how can I predict positions that do not appear in truth? Because I don't have information about this positions (can't use in-situ matrix). Wow! So with the in-situs file (bdtnp.txt for raw data and binarized_bdtnp.csv for binarized data), we can only be used to calculate the actual position (DistMap) and get in-situs genes names. Addition, we can't use it for anything else (e.g get sub-in-situ matrix 20/40/60 genes). Maybe, I've misunderstood before.
That is correct! Sorry, but I just want to confirm that _choose 20/40/60 genes_ means I only get names of genes and can't get information of genes (vector binary) in the in-situ matrix. Let me put it this way"
1) You can use all RNAseq data and cell positions as given by DistMap
2) You can only choose your 20/40/60 gene insitus once and THEN do the learning. You cannot subselect 20/40/60 in situs performing selection on the 84 in situs
Drop files to upload
Supervised learning or Unsupervised learning ? page is loading…