In the spirit of getting a beautiful solution applicable to many types of protein behavior, we've been speaking pretty broadly about local and global behaviors. However to actually characterize our mitochondria data, we only need to recognize fission and fusion, as actions of the cell rather than classes representing independent states. Mito. behaviour is a compound action of these two actions as well as shape changes in the cell membrane.
The data for control, mdivi, and llo can be broken into:
Presence of fission:
LLO and control
Presence of fusion:
mdivi and control, and the beginning of LLO
So could build a simple action recognizer to classify a chunk of video frames into fission or no fission, and vice versa. Would end up with interpretable results -> instead of anomaly detection where a dist. of protein simply doesn't fit into the expected pattern, diseased cells can be characterized by the imbalance of these two actions and/or drastic shape change.
Fission and fusion in turn have a few components to capture - mass change (protein increase/decrease) and topology change(mass stays the same but protein connections are redistributed). Drastic shape change seems to be a result of imbalance in fusion/fission, like a cellular event that is triggered.
In the spirit of getting a beautiful solution applicable to many types of protein behavior, we've been speaking pretty broadly about local and global behaviors. However to actually characterize our mitochondria data, we only need to recognize fission and fusion, as actions of the cell rather than classes representing independent states. Mito. behaviour is a compound action of these two actions as well as shape changes in the cell membrane.
The data for control, mdivi, and llo can be broken into: Presence of fission: LLO and control Presence of fusion: mdivi and control, and the beginning of LLO
So could build a simple action recognizer to classify a chunk of video frames into fission or no fission, and vice versa. Would end up with interpretable results -> instead of anomaly detection where a dist. of protein simply doesn't fit into the expected pattern, diseased cells can be characterized by the imbalance of these two actions and/or drastic shape change.
Fission and fusion in turn have a few components to capture - mass change (protein increase/decrease) and topology change(mass stays the same but protein connections are redistributed). Drastic shape change seems to be a result of imbalance in fusion/fission, like a cellular event that is triggered.