Using a verification model on multiple instruments expands a manufacturer’s raw material identification/verification (RMID) capabilities by speeding up incoming inspection, imparting flexibility to an operation, or avoiding downtime.
In a scenario where several operators use multiple MIRA P systems at different locations, the ability of any operator to use any MIRA P to validate a new shipment streamlines operations and allows that shipment to be quickly released to production.
In most cases, a well-designed model with inherent sample variability can be built on one MIRA P and transferred to another. In some cases, variance must be added to a training set with a few additional samples. This Application Note describes how a model transfers from one MIRA P to another in order to scale MIRA P usage across an entire operation.
Model building (including training and validation set samples, operating procedure (OP) settings, and necessary variance) has already been well-established for RMID with a unique MIRA P [1,2].
In summary, MIRA Cal P generates PCA-based (principal component analysis) models using Training Set data and Operating Set parameters to verify target substances. Ideally, a model can be created on one instrument («MIRA P 1»), downloaded onto a second instrument («MIRA P 2»), then validated on the second unit and used directly.
The model must be expanded if the initial transfer does not produce satisfactory p-values or does not pass validation. This involves introducing variance in the model and/or optimizing model parameters and/or ensuring consistent usage by each operator of the instrument.
This Application Note details the:
Contact your local Metrohm Sales and/or Service Representative for the full MIRA P to MIRA P Transfer Protocol.
Parameter optimization and/or inclusion of additional data that includes instrument variance and variance based on historical and current samples are both simple ways to expand a model.
Using an established model, new validation data is collected from both MIRA P units and added to the Training Set. Parameter optimization is recommended at this step. After this updated model is uploaded to the new MIRA P unit, it must be validated on the new unit.
Validation of a model demonstrates that the model adequately assesses a material on a new instrument. In other words, validation data serves as a «diagnosis» of how the model performs on the new unit.
Validation is an assessment of a method using test samples:
It is a simple task, requiring just a few minutes, to run a validation set. This will inform successive steps.
A good way to assess the success of a transferred model is to look at p-value distributions of positive and negative Validation Set samples. This is a good measure of a model’s robustness—its ability to correctly assess new data, not just the data it was trained on. For example, Figure 1 contains Validation Set results for sodium bicarbonate on the receiving MIRA P device (MIRA P 2).
Initial Validation results for sodium bicarbonate show all the characteristics of good validation results (Figure 1a).
Red bars indicate that negative validation samples are failing appropriately, and positive samples are also passing with high p-values.
After transfer to MIRA P 2 (Figure 1b), p-values for negative and positive samples show greater variance, but all are passing/failing appropriately.
Ultimately, this is a good example of a model that was transferred and used immediately.
Lactose fluoresces with 785 nm Raman, but a well-built model can accommodate such fluorescence. This is a good test of the model transfer capability.
The lactose model transfers easily, requiring only parameter optimization in MIRA Cal P and addition of a small number of scans from the new instrument to the training set (Figure 2).
P-values exhibited slightly more variance on the new instrument (Figure 2b), but the model was robust.
Microcrystalline cellulose (MCC) is a challenging sample for 785 nm Raman, as it is very fluorescent. This can be seen in the wider distribution of validation set p-values in the original model (Figure 3a).
Thus, it was not expected for the original model to transfer without the Transfer Matrix.
Ultimately, the optimization of parameters and utilization of the Transfer Matrix provided a robust model that could be used on a second MIRA P instrument and with a smaller spread of p-values.
Compliance with 21 CFR Part 11 requires document control at the highest level. Specifically, all model metadata is preserved, ensuring traceability after model transfer. The manufacturer-supported means for this are simple: MIRA P Model Transfer Protocol includes a sign-off sheet to record and track data from MIRA Cal P export, through the transfer matrix, and import back into MIRA Cal P.
The benefits of using multiple MIRA P devices for raw material verification include smoother operations and faster turnaround of products. This Application Note is intended to guide users through model transfer and enable the deployment of multiple MIRA P instruments.
From tips for the simplest transfer to tools for more challenging tasks, we want you to be confident in taking your inspection with MIRA P to the next level. This flowchart is a quick reference for the basic flow of operations during MIRA P to MIRA P transfer.
Chia sẻ ứng dụng
Tải về bản PDF