Moderated by Stephen Muggleton.

Jacques Cohen

Classification of approaches used to study cell regulation: Search for a unified view using constraints and machine learning

The article mentioned above has been submitted to the Electronic Transactions on Artificial Intelligence, and the present page contains the review discussion. Click here for more explanations and for the webpage of theauthor, Jacques Cohen.

Overview of interactions

N:o Question Answer(s) Continued discussion
1 15.10  Anonymous Referee 1
2 15.10  Anonymous Referee 2

Q1. Anonymous Referee 1 (15.10):

Overview of the paper:

Modeling by differential equations had been popular for the simulation of cell regulation. After the popular use of micro array, graphical modeling became popular. Prediction of the cell regulation by simulation is a direct problem and the modeling is a inverse problem. The approaches for those problems are classified by the following three viewpoints:

Probabilistic versus Non probabilistic
Continuous versus Discrete
Direct versus Inverse

In this paper, methods of simulation and modeling of expression patterns of genes are classified from the viewpoint of machine learning.


Recommendation: The paper can be accepted for ETAI after minor revisions. No further refereeing round is needed.

In order to simulate the cell regulation, it looks sufficient to select appropriate model for each case. The necessity for united interpretation using several assumptions is not clear. In the modeling problem, it is shown that the space of variable vectors can be divided into subspaces. The problem is simplified by considering mutual transitions between those spaces. That can be a good approach for the modeling problem.

Detailed comments are follows:

2 The Direct Continuos Approach
It is acceptable that sigmoid function is adopted as an example of continuous modeling. However, it may not be appropriate to use the viewpoint of automaton, because the steepness of the sigmoid function is ignored. It is recommended to explain how to treat time in the example of Figure 3. In Figure 1, x_i should be f_ij.

3.1 The Inverse Problem
When the mutual interactions of each variables have the type of sigmoid function, is it sufficient to give the trajectory to specify the constraints? In that case, is it possible to select small number of graph structures from numerous number of the structures ?

5. The Qualitative Approach
It is not always sufficient to approximate the set of differential equations by step functions. How are they treated when they increase/decrease in linear manner?

It is recommended to show the case which has three or more state variables.

8 Final Remarks
It is recommended to state about the application of the proposed methods to the expression profiles after the clustering.

Q2. Anonymous Referee 2 (15.10):

Overall advice: The paper can possibly be accepted after major revisions and another round of reviewing.

This advice comes from an AI scientist with little background in the specific problem addressed here. So, it is a low confidence review.

Motivation: the work is quite interesting, but still preliminary It describes a problem (the inverse problem) but does not really present solutions as yet. It is more like a challenge Also, there is neither real data nor any experiments taht backup the sketched approach. Additionally, not all techniques that seem potentially applicable to this problem are discussed (e.g. neural nets). Finally, the part on cell regulation (and some background on cell regulation) should be expanded for an AI audience. Not everybody is familiar with that.

In my printed version of the paper, there were some minor problems with the figures (arrows not always present), but it is unclear whether this is due to my printer or to some other cause

Minor comments:

P1. several papers are now available : please name them

P2. and elsewhere : the author refers a lot to ILP. This is not the only form of mahcine learning, and other forms of machine learning, such as neural nets, should also be discussed.

P3. I assume xi denotes the expression level of hte gene ?

P4. section 2.1. please expand. One might also use equation generation systems such as those by Washio and Motoda

P6. the inverse problem. It would be good to specify the problem more formally, and to illustrate it with a detailed example.

The inverse boolean approach is quite clear. It would be good to give as detailed an example for the other approaches too. Given the inverse boolean problem, I am wondering whether it is not possible to use (recurrent ?) neural networks for addressing this inverse problem as well as the other ones ?

P15. I do not really agree with the (2) point and its clarification. In fibonaci, you can indeed use the program in two directions. It seems however that for the clps given in hte earlier sections, the goal of the inverse problem is to construct parts of the program (like the slopes etc). I do not really see the analogy there. It would be good to clarify this further on a detailed example in the earlier sections.


Background: Review Protocol Pages and the ETAI

This Review Protocol Page (RPP) is a part of the webpage structure for the Electronic Transactions on Artificial Intelligence, or ETAI. The ETAI is an electronic journal that uses the Internet medium not merely for distributing the articles, but also for a novel, two-stage review procedure. The first review phase is open and allows the peer community to ask questions to the author and to create a discussion about the contribution. The second phase - called refereeing in the ETAI - is like conventional journal refereeing except that the major part of the required feedback is supposed to have occurred already in the first, review phase.

The referees make a recommendation whether the article is to be accepted or declined, as usual. The article and the discussion remain on-line regardless of whether the article was accepted or not. Additional questions and discussion after the acceptance decision are welcomed.

The Review Protocol Page is used as a working structure for the entire reviewing process. During the first (review) phase it accumulates the successive debate contributions. If the referees make specific comments about the article in the refereeing phase, then those comments are posted on the RPP as well, but without indicating the identity of the referee. (In many cases the referees may return simply an " accept" or " decline" recommendation, namely if sufficient feedback has been obtained already in the review phase).