tnaake / metnet Goto Github PK

• only use msLevel 2 from a spectra object
• match against the rownames of a AdjacencyMatrix object
• calculate the MS2 similarity based on a specified score
• add assay slot MS2

Currently combineStructuralStatistical accepts two matrices (one structural one statistical inferrence).

• write the function that it accepts the list of matrices from createStructuralAdjacency
• return a list of two matrices where the second matrix contains the type with removed types that are not supported by the statistical adjacency matrix

when the m/z differences between two features is small it is possible that one of the differences is negative while the other is positive (see the photo attached).

• check: is this only a problem in directed = FALSE (as we check for the interval in directed = TRUE, without taking the absolute values)
• when we take the absolute values of the differences, we mess up with the negatively signed difference, e.g. -1.12 will be converted to 1.12 -> we will then check in a interval of 1.12 and the upper bound. FIX: set the 1.12 to 0 when there are elements of different sign at the same corresponding cell.

@elvanov could you have a look into that and suggest a bug fix (if you do it your account is connected to the repo). The added code could go into (before the mat_1_abs <- abs(mat_1)

    if (!directed) {
        mat_1_abs <- abs(mat_1)
        mat_2_abs <- abs(mat_2)
        mat_1 <- ifelse(mat_1_abs <= mat_2_abs, mat_2_abs, mat_1_abs) ## max
        mat_2 <- ifelse(mat_1_abs > mat_2_abs, mat_2_abs, mat_1_abs) ## min
    }

for the function structural, add the possibility to add customized assays that are specified by the user, e.g. add an assay containing the ID of the type of transformation (e.g. EC Numbers or Inchi keys, ...)

• implement in combine
• roll over the assays in combine

• return the raw value, do not return binary values as now, but create this only later
• top 1: take top N from all methods, can return more than N features
• top 2: take 2nd top N from all methods, returns max N features
• mean: take the averaged from all methods top N: returns max N features
• threshold: define for all methods threshold and return values that are above the threshold

MetNet currently depends on 138 dependencies. Try to reduce the number of dependencies.

Implement sparce matrices to speed up and reduce memory load

• rename estimate to r for partial and semipartial
• rename p.value to p for partial and semipartial

when the peaklist contains MVs semipartial/partial correlation will fail

• write unit tests for this case
• set NA for coefficient and p-values for feature pairs where features contain MVs