An Unambiguous Algorithm

The algorithm of an acceptable QSAR model, that correlates the studied endpoint with chemical structure using molecular descriptors, must be reproducible and easily applicable, even by non-experts. For this reason, there is a 

The QSAR algorithm establishes correlation between the studied response and the molecular descriptors, but recently some concerns have been raised in the lit-erature, emphasizing that correlation between variables does not automatically imply that one causes the other and that chance correlation could occur, mainly if not understandable descriptors are used. However, correlation is a fundamental requirement for causation. The best way to exclude chance correlation is to carefully verify the statistical predictivity of the QSAR models by their validation, as requested by OECD principle 4, also externally on new chemicals, and by scrambling of the response, additionally, if possible, to mechanistically interpret the molecular descriptor (OECD Principle 5). If the correlation is confirmed after rigorous verification, it has a reason for existing and the problem is one of the human mind if the cause is not discovered or understood. 

An OECD (Q)SAR application toolbox for REACH has been developed and implemented to categorize chemicals by different tools (for example Toxtree, Toxmatch, in Joint Research Center (JRC)-QSAR tools ) and to predict toxicity mainly by read-across techniques. This is suggested for regulatory use. 

---

The most important one is that the model should be appropriately validated to confirm the reliability of its predictions. First rules of the validation were worked out in March 2002 at an international workshop held in Setubal, Portugal ( Setubal Rules ). In November 2004, the rules were discussed and modified by the OECD Work Program on QSAR they are now known as the OECD Principles. According to these principles, each QSAR model should be associated with (a) a well-defined endpoint (b) an unambiguous algorithm (c) a defined domain of applicability (d) appropriate measures of goodness-of-fit, robustness and predic-tivity and (v) a mechanistic interpretation, if possible [15, 16]. 

In general, the rules pay a close attention to transparency when the model and the results are presented. Not only well described data, but also an unambiguous algorithm must allow everybody to be able to repeat the modeling. Therefore, the model cannot be reported as a black box, but all necessary details on the mathematical method, descriptors, etc. should be given in the report [15,16]. Currently, the universal... 

An unambiguous algorithm This is to ensure transparency in the model algorithm. Without this information, the performance of a model cannot be independently established, which is likely to represent a barrier for regulatory acceptance. 

An unambiguous algorithm. The intent of this principle is to ensure the transparency of the modeling algorithm. Sometimes, it is a difficult task to satisfy this principle, particularly when complex methods like neural networks or fuzzy logic techniques are used for modeling. 

An unambiguous algorithm. It is a necessary to use an unambiguous algorithm for the model building ... 

Principle 2 an unambiguous algorithm the Spectral-, Diagonal-, Projective-, Quantum Amplitude-, Residual-, Alert-, Catastrophe-, SMILES- and topo-reactive QSAR variants ... 

Figure 2-42. The Morgan Algorithm generates an unambiguous and unique numbering of phenylalanine (see Tutorial, Section 2,S,J,1).

The constitution can be represented in an unambiguous and unique manner by canonicalization (Morgan Algorithm). 

The a priori localized nature of geminals is quite important and promising. Computationally, localization permits one to develop efficient algorithms that scale, in the limit of large Ny quadratically or even linearly with the number of geminals (N). Conceptually, localization means that a geminal-type wave function may represent classical chemical concepts (such as the two-electron bond, lone pairs, etc.) in an unambiguous manner. In fact, we believe that the best possible quantum chemical representation of local bonds in a molecule is an APG-type wave function. Thus the limitations of this model reflect the limitations of these concepts themselves. 

A special extension of SMILES is USMILES (sometimes described as Broad SMILES) [23-25]. This Unique SMILES of Daylight is a canonical representation of a structure. This means that the coding is independent of the internal atomic numbering and results always in the same canonical, unambiguous, and unique description of the compound, granted by an algorithm (see Section 2.5.2). 

---