Julia's awesome untitled project

[neophilology]

Hi @juliaquach02

Great explanation. From the intro to the goals, all was clear, descriptive, and succinct. The pictures and plots were extra helpful and a nice touch. No working title yet?

The data is fantastic and straightforward. Do you know how the measurements were collected, if by "hand" or computer vision?

I have some comments regarding the goals, which I'll address point by point:

"Identify dying tumor cells in the data sets and estimate the point of death."

- I am thinking we could establish a "breaking point" or size-fluorescence threshold. We could use/impose our standards or use an ML algorithm for unclassified data. Do you have something in mind? (This is interesting, I'll review it more closely in the following days and come up with something by Tuesday.)
- After this step, we could add the "time of death" (time-series or interval?) and labels (maybe a binary alive or death or dummy variable) as new columns. 

"Count dying tumor cells and compute the relative number of cell deaths."**

- This will be manageable after having the new values in the data frame.

"Plot the cell survival for both cell populations in one plot."

- I have to confess that I'm having trouble plotting things in R. This will be the most challenging part for me, but I'm here to learn, so I will ;)
- Maybe we could also plot a comparative chart of the treatments' efficacy (how fast they shrink and die).

One last point to think about: Cell location: the x and y columns. Why do we need them if "shrinking size and fading fluorescence intensity [are the] two necessary criteria"? Do you consider these columns could add information when classifying cell death rate and their time of death? I worry they could be irrelevant to the study you want and even add noise.

[martin-raden]

Hi both, thanks for the nice project and the great review. Both looks good!

Here some points from my side, how you might improve the project description.

• [x] try to shrink the images a bit (or replace by smaller cut-outs). they take a lot of space but are (in the end) "only" important for the story. Alternatively: put them side-by-side in one row to save space.
• [x] Is the length of a cell's data set (number of time points) already sufficient to say whether it is dying or not?
• [x] What is a "measure" to say a cell is dying? linear downward trend (what angle is a threshold?)? end area size only x% of start size? ... I think some information in that regard are important to sharpen the goal. Otherwise, you will get a lot of questions.. ;)
• [x] It's a lot of data. You might want to suggest to first start with a minor subset, before extrapolating to the whole data set. Maybe suggest a subset that includes some dying and non-dying cells given your expert knowledge.
• [x] Is the "finally killing stroke" via paraformaldehyde at a specific time point? If so, specify. If not, given an idea how to estimate..
• [x] Clearify that the target plot will only contain TWO lines (in contrast to the provided example plot)

Great project. Looking forward to see it solved!

Best, Martin

[martin-raden]

see also #114

[juliaquach02]

Hi @neophilology and @martin-raden,

thank you for your kind and thoughtful feedback!

@neophilology

I will address your feedback down below:

No working title yet?

My working title is “Investigating tumour cell viability” but apparently, the name only appears in the list of projects but not under the link itself. I will ask on Tuesday how to fix this.

“Do you know how the measurements were collected, if by "hand" or computer vision?”

That is a good question. The measurements were collected using a convolutional network called Stardist to identify and delineate cells and a tool called TrackMate to track cells over time. I added links for further information in the project description.

“I am thinking we could establish a "breaking point" or size-fluorescence threshold. We could use/impose our standards or use an ML algorithm for unclassified data. Do you have something in mind? (This is interesting, I'll review it more closely in the following days and come up with something by Tuesday.)”

I was also thinking about introducing thresholds regarding the cell size and fluorescence intensity. To evaluate the cell size, we could view all cells for which the cell size drop by (let’s say) 40% as dying. For this criterium, I added additional information in the project description.

For the fluorescence intensity, it is a little trickier as the fluorescence intensity of all cells might fade over time due to bleaching effects of the microscopy lamp. I checked the fluorescence intensity of a couple dying cells. For those dying cells, I found a sudden, clear linear decline in fluorescence intensity while for the viable cells, the fluorescence intensity fluctuated. Maybe this could be a criterium? Otherwise, we could use a fluorescence intensity threshold relative to the maximum fluorescence intensity value of the cell (maybe also 40%).

Using an ML algorithm for unclassified data sounds exciting! I am not familiar with ML algorithms, but you are very welcome to test them on my data.

Maybe we could also plot a comparative chart of the treatments' efficacy (how fast they shrink and die).

Great idea! In theory, this would be awesome, but I fear that the data sets are too messy to compare these aspects of cell death.

Why do we need them if "shrinking size and fading fluorescence intensity [are the] two necessary criteria"?

You are absolutely right. The coordinates of the cells are not necessary for classifying cell death and time of death. I removed the columns from the data set to avoid confusion. Thanks for this remark!

@martin-raden Thanks for your helpful comments! I implemented your feedback and added additional information regarding your questions in the project description.

Best Julia