Do I need to filter out low abundance MS1 features i.e. those with low signal/noise to control for outliers so normalisation works?

No. We use the median and also the median absolute deviation as an approximation of the variance to remove the influence of outliers. We iterate this procedure to determine statistical bounds for outliers so these outliers have no effect on the result of the normalisation calculation. This means there is no need to filter out data at the MS1 level and means you can work with all the data that is available to you. This would include low abundance compounds that may not be indentified but are of biological significance and can be followed up with de-novo sequencing approaches.