Analyzing Biomedical Datasets with Symbolic Tree Adaptive Resonance Theory

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

The paper was really improved with respect to the previous version. However, points of weakness remain that need to be further improved. 

I insist about Algorithm 1.  The authors cannot identify an algorithm with a "Shared START notation". A notation is not an algorithm, at the same manner that a function is not an equation or a number is not an operation. In a sciebtific paper such kinds of terminology abuses are not acceptable.. Therefore, please replace  "Algorithm 1" with  the appropriate title.

About the algorithms, their description is really very hard to follow, just as an example. In the following fragment taken from Algorithm 4, for the reader is almost impossible to follow the passages where symbols occur that can be firmly understood only by the author of the code. I sugest to simplify the statements, leaving only the informal description of the action performed by the instruction. An assignment of variables that are not easily meaningful is completely useless.

===================

 ù/* Create and initialize new category */

2 K ← ∥C∥1 + 1

3 RK ← fN(x,G)

/* Add the label to the label map */

4 UK ← ω

===============================

 

 

About the algorithms, it would be very useful to know how these algoritms were really implemented, in wich language, by providing also technical details about the hardware and their performace.

 

Figure 1 and Figure 2 are not figures, but tables. Please use the correct terminology.

 

As already noticed in my previous review, more examples would be very useful. For example, the reader would see some real statements describing  clinical data of the considered disease, or a comparison of different statements related to different clinical situations, by observing what really changes in the linguistic reprersentations of different cases.

 

Finally,  all discussion seems yo be autoreferential, because it is not clearly expressed if the analysis of the considered disease really gave some clinical advantage with respect to the traditional clinical approaches. What was really discovered that could be considered an advancement in the treatment of the disease?

 

I hope that authors can improve the paper by removing the mentioned limits, and stressing the points of interest of their results.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

Improvements have been made

Author Response

Please see the attachment.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper presents  a continuation of ART (Adaptive Resonance Theory) and an application to the analysis of a disease dataset.

The paper is surely related to an important research activity, but in many points it needs clarifications for a wide audience comprehension.

The paper is well written and organized, but it assumes that the reader knows completely the field of the subject. The algorithms are described in their general schema, but the real advantages os the design choices are not sufficiently motivated.

The authors have  to communicate the main ideas of their research in a way that is indetendent of a knowledge of the papers to which it relies.

Let me give an example from the section 2.2 (Gram-Art),

Lines 110-117.

In the original formulation, Gram-ART samples  are statements adhering to a CFG that are parsed into rooted syntax trees. These parsed  samples are then compared according to ART learning rules to Gram-ART prototypes that are themselves rooted trees  ***containing distributions of encountered terminal symbols at  each node***. Gram-ART answers the questions of how to formulate prototype trees of varied shape, compute similarities of sample statements to prototypes of differing shapes, and 

update the terminal symbol distributions at each node during learning.

 

The expression between *** is speaking of distributions of terminal nodes “encountered” at each node. What doet this mean? Encountered where? During the parsing of the statements? In which sense? And how prototypes express the target of the learning process toward the required clustering? The absence of examples (even taken from the disease considered in the paper) does not help the reader to understand the essence of the proposed algorithms.

I suggest the authors to give examples, even toy examples, of the things the paper discuss; statements, grammars, trees, prototype trees, distributions, and so on, in order to help the reader to understand the logic within which their proposal can be evaluated. 

The same criticism applies to many aspects that are mentioned, but without no specific reference, apart from bibliographic references, which are of course important, but  cannot replace the basic descriptional level of a paper.

About Algorithm 1: Where is the algorithm?

About Algorithm 2: Data section reports: “S Symbolic Statements S” and at the same time, ” S as Statement definition” (what is?).

Which kind of activation function are used, and what does it mean the activation of  a node. It is not clear if the competition is among nodes of the same tree or something else. Without a clear comprehension of these nain points, Figure 3 does not help.

Maybe, when key concepts and procedures are clearly stated, the dscussion after the algorithms can be easily followed, but again, the main ideas of ART and its notation cannot be devolved to the quotation of the original work, because the reader who is not familiar with the subject cannot gain a real knowledge from the reading of the paper.

In the case of the considered disease, what are the items on which the algorithm work in the different categories (statements, trees, ….)?

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

-an illustrative example using a diagram for the START operations included in page5, lines:193-199 could be useful for the readership.

-a table with the four algorithms including their characteristics, differences, and advantages could be more informative.  

-it would be useful to show if there is a preference of a certain algorithm to be used for  clustering depending on the type of a dataset.

-in the experiments authors should state which of the proposed algorithms has been applied.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf