Developing & Automating Pipetting with Pipetting Robots

Andrew Alliance has designed and optimized a procedure for assessing the performances of pipette performance in a normal laboratory environment with the practical objective of a reproducible process without subjective interpretations.

In contrast to many present calibration methods that necessitate highly skilled operators, Andrew Alliance aims to produce a procedure that can be executed by anybody in the lab, automatically, and without the need for expensive equipment and infrastructure to achieve narrow control of environmental condition.

Additionally, the company aims to demonstrate a procedure that can measure a complete set of pipettes without the need for oversight. This enables users to verify the consistency of performances within the full range (in this case, from 0.2 µL to 1000 µL) of liquid handling dynamic, not including gaps and measuring the performances in precisely the same conditions of real life experiments.

The experimental setup is made up of the Andrew robot manipulating the Gilson Pipetman Classic pipettes P2, P20, P100, P200, and P1000. This choice of pipette models and type, considered amongst the most performing among all manual pipettes, enabled the full coverage of the range 0.2 to 1000 µL, which is mostly used in conventional biology.

The Andrew Robot

Andrew is a compact, 10 kg robot boasting easy installation onto any lab bench (only taking up as much space as a sheet of paper) and simple transportation. Despite this, the Andrew robot has the capacity to perform both large and small experiments, for example from 1 to 11 microplates or from 1 to 165 microtubes.

Based on vision and advanced software algorithms, the robot can identify consumables, their location, and operate the pipettes exactly as human users would do. Andrew can take a pipette, set its volume, insert and eject tips, aspirate and dispense precisely and reproducibly without any user intervention and with precise timing and operational speed.

The Andrew robot utilized for this series of experiments is a commercial model (Andrew#132 when not otherwise stated) without any modification to hardware or software, using AndrewOS version

The light-absorbing liquid samples used consist of solutions containing Ponceau S obtained from Sigma Aldrich. These solutions were diluted into analytical grade water, but results were also reported with solution using additional detergents or other additives to achieve different viscosities.

In fact, it is well known, that the surface tension properties and the viscosity of liquids strongly affect the pipette performances, and the described setup allows for the reproducible measurement of the pipetting performances with any liquid, without the subjective component induced by a human operator manipulating the pipettes.

The pipette-produced samples are measured by an Andrew Alliance photometer or by a reference balance.

The choice of a custom photometric reader means Andrew Alliance can precisely fix the optical conditions and internal calibrations. The Andrew Alliance absorbance reader is a compact, LED-based photometer made for reading clear-bottom 96 wells microplates. It measures the optical absorbance of samples at a 520 nm wavelength with a calibrated light beam that crosses the sample from bottom to top. The passing-through light beam intensity is measured by a photodetector that can linearly cover the range of Optical Density from 0 OD to 4 OD.

The reader includes shaking of the microplate during readout to avoid problems with mixing and any effects related to the meniscus shape, which complements sample mixing achieved by multiple aspiration and dispensing actions. Furthermore, the reader also integrates an internal calibration method to gather reproducible results without the need for external optical samples with known absorption properties.

The photometric method used consists of dispensing dye-loaded samples with a known concentration into a clear-bottom microplate, where the microplate, reader and liquid sample properties have all been pre-calibrated by recording the samples’ absorbance in these optical conditions as a function of their weight. This means that Andrew Alliance can indirectly exploit a gravimetric method in order to calibrate the photometry.

A key benefit of this particular method is in achieving a calibration measurement in an unattended manner. Boasting a short turnaround that can be effectively carried out in the same lab.

Andrew Alliance exploits the same protocols for gravimetric and photometric measurements and the same pipetting procedures. The specific volumes used for each pipette are reported in the table below. Ten repetitions are performed for each volume, resulting in three different volumes for every pipette.

Due to the overlap of the pipette ranges, only the volume range actually used by Andrew is measured. For better clarity, the pipette P200 is tested in the range 100 µL to 200 µL as this is the range used during operations employing the use of the Andrew robot.

Andrew Alliance does not probe the 1000 µL volume corresponding to the maximum volume of the P1000. This is due to the limitation of the maximum volume inside a clear-bottom microplate of the order of 350 µL. However, this limitation can be avoided by using a clear-bottom 24-well microplate that is compatible with the reader.

For the volumes below 200 µL, the sample is introduced into a buffer-containing well for an equalization of the optical conditions. This also prevents evaporation in the sample for the smallest volume range. As such, all final samples contain 200 µL of liquid with the exception of the 300 µL calibration for the P1000 pipette. As discussed in the next section, verification of the amount of buffer was possible, and its intrinsic optical properties do not affect the measurement conclusions.

For each of the above-mentioned volumes, Andrew Alliance has prepared 10 independent wells in the previously described manner for a total of 150 sample readings over two microplates. The remaining wells are filled with additional controls, in particular blanks, using all 192 positions in two microplates.

The microplate reading only takes a number of seconds after thorough mixing, which is achieved by aspiration and dispensing with a large volume tip, an operation that is performed automatically and with high reproducibility by the robot itself.

Regarding gravimetric methods, Andrew Alliance exploits the conventional guidelines for the use of a semi-microbalance. As a reference balance, they have opted a Mettler Toledo AX205DR (8). There’s no commercial balance covering the full dynamic range of required volumes with the necessary precision and accuracy, an issue that is important to highlight.

In fact, the low-volume range (below 2 µL) puts extremely stringent requirements on the balance performances, requiring ultra-precision balances capable of measuring samples with a reproducibility that should be better than 10 micrograms.

The balance can become extremely costly and the environment conditions become challenging, requiring a constant temperature within 0.5 °C, high humidity levels of over 50%, critical insulation from vibrations through marbles, a minimal distance from vibration sources like traffic, and control of the air draughts in the surroundings. The company was aware that the comparison between the two methods should be performed in real life laboratory conditions, and so used a single, semi-microbalance with a nominal weight resolution with the economical equivalent to a photometer, and it was used with the reasonable precautions in a normal laboratory environment.

Andrew followed all pipetting guidelines suggested by the pipette manufacturer regarding the correct use of the tools in every pipetting operation. Profiting from the reproducibility the Andrew robot offers, the company has individually explored and studied most critical aspects affecting liquid handling performances, confirming our conclusions in the existing literature (see, for example, an interesting independent pipetting review in (9)). However, the work has revealed some subtle clarifications on a specific procedure that enables users to substantially enhances pipetting performances. This particular procedure is tip pre-wetting.

Pre-Wetting: There’s More To It

The majority of users are aware that to dispense the correct volume, pipette tips have to be pre-wetted. However, this practice is often neglected. This is because it suggests a significant workload that will add time onto experiments and increase the chances of mistakes occurring such as carrying out the incorrect sequence of steps or the wrong localization of wells into the microplate.

It is interesting to note that the role of pre-wetting is usually emphasized in the context of liquid adhesion to the tip surface, which improves and stabilizes the contact angle of the liquid inside the tip. While it is an evident effect, in particular with some liquids having a high affinity to polypropylene, it is not the only phenomenon occurring during liquid aspiration.

Andrew Alliance has experimentally proved that there is a more important reason for performing pre-wetting, especially when the environment is characterized by the normal humidity and temperature states seen in a real life lab: vapor build-up. This understanding impacts the pre-wetting procedure and can improve the pipetting performances by avoiding systematic errors that can easily achieve 1 µL for a volume of 100 µL (1% systematic error): as it can be seen in Table 1, the magnitude of this systematic error induced by the method is by far larger than the admitted tolerances indicated in the ISO norms. Consequently, when pipetting accuracy is important, following the procedure described below is suggested.

Andrew Alliance discovered that the majority of users carry out pre-wetting by putting the tip into the source and aspirating the liquid inside the tip a number of times. After this step, the sample is usually aspirated inside the tip without extracting it from the liquid. However, Andrew Alliance was able to determine that this procedure doesn’t enable users to achieve the best pipetting performance.

There is a second, arguably more important role to pre-wetting that is often overlooked. Pre-wetting also increases the air humidity inside the tip and the piston cavity in order to achieve a stable vapor pressure.

Once a liquid is aspirated into the cavity of the tip and the piston for the first time, the surface of the liquid, since coming into contact with the dry air, continues to evaporate. Molecules of water develop into their gaseous form to reach equilibrium, which shifts the liquid down by the volume equivalent to the molecules that underwent phase change transition.

To reinforce this fact, consider that the boiling point 1 nL of water becomes 1.9 µL of vapor. This shows that minute amounts of evaporated water can have a substantial effect on the volume of liquid retained inside the tip of the pipette.

Carrying out numerous aspiration and dispensing actions on the pipette plunger performed by keeping the tip inside the liquid bulk are generally not capable of expelling the extra volume of water vapor and air inside the tip because of surface tension effects. In consequence, the actual volume of liquid aspirated from the sample without removing the tip from the liquid is reduced correspondingly.

In order to deal with this phenomenon correctly, it is enough to extract the pipette tip from the source after pre-wetting, and re-insert the tip into the liquid for the final sample aspiration before dispensing. By doing this, the tip and piston cavity is correctly saturated with vapor and the amount of liquid aspirated into the tip corresponds to the nominal volume of gas (air and water vapor) displaced by the pipette piston.

The atmosphere inside the tip is not perturbed by the extraction in open air due to the impedance of the tip orifice. Performing this procedure correctly with Andrew, it has been possible to precisely and repeatedly achieve the nominal performances of the pipettes used.

Experimental Data

In photometric experiments, the instrument background has been subtracted by direct measurement from blanks (water). After this, absorbance is converted into actual microliters according to a single gravimetric assessment. As a result, random errors are calculated based on the standard deviation of the samples, irrespective of the calibration scale.

The systematic error of each volume set is determined on the basis of the nominal dispensed volume and the average value measured on all the samples. In every instance, the calibration scale is created by a single reference absorbance measurement of a weighted sample that has been read in matching conditions, allowing for the conversion of absorbance data into microliters. The tables below report typical data gathered by this photometric method:

These results are reported in a chart normalized to the ISO-8655 upper limits, both for systematic and random errors for every single measurement. In this plot, a systematic (random) error is equal to 100% when it corresponds to the ISO systematic (random) error limit, and is proportionally below if the systematic (random) error is smaller.

It is also interesting to note that random errors are entirely positive, while systematic errors can be either positive or negative. It is clear that all measurements do remain in the required accuracy limits, but the larger volume measurements approach the allowed upper values.

Additionally, data was gathered in the same experimental configuration with the gravimetric method. The protocol for the gravimetric method is essentially the same as the photometric method, but the liquids are dispensed into a conventional microtube of 1.5 mL volume.

The Andrew robot stops after every dispensing action to put the closed tube onto the balance. This procedure allowed for the measurement of the source and destination tube for each individual measurement at the same time, generating information about the correlation between the volumes of aspirated and dispensed liquid.

As expected qualitatively, the photometric and gravimetric method differ quantitatively in determining the ultimate performance of pipettes. Highlighted below are explanations for this behavior, explanations that can be placed into the two general categories:

  • Experimental resolution of the measurement instruments: reproducibility and calibration (linearity)
  • An intrinsically different sensitivity to dye losses/mixing for photometry vs. sensitivity to buffer losses/evaporation for the gravimetric method

To close the final chapter a critical review is conducted, of the data after the systematic review of possible random and systematic sources of errors, and the comparison with other liquid handling conditions and procedures.

Analysis of Intrinsic Measurement Errors

This section systematically reviews all intrinsic sources of systematic and random errors that could affect the final data.

Photometric Method Errors

Same Well Absorbance Readings (Statistical Reproducibility)

The most obvious source of error actually comes from the absorbance reader itself. A range of potential issues arise in regards to the light source technology, light detection technology, software algorithms and reading method.

These uncertainties can’t simply be extrapolated from the optical reader manufacturer specifications as the reader specifications often allude to conditions that differ from our own measurements.

For example, noise is strictly a function of the absorbance and is variable across the range of measurement. Therefore, attention was turned to the most basic measurement: assessing the reproducibility for reading a single well (the same well with the same sample) multiple times. The result, in particular the relative standard deviation (CV) based on 30 consecutive readings, is reported in the plot below as a function of the absorbance:

It is clear that there are two regimes to avoid: the case of samples with low concentration/low absorbance (left part, where the error rises steeply for absorbance <0.1 OD) and those where the sample absorbs most of the light (right part, smooth error increase beyond ~3 OD).

To get random errors introduced by the reader, all experiments need to be designed in order to produce data that is limited to the useful range. All data discussed in this article has been limited to the 0.3-3.0 OD range in absorbance. In this range, the statistical error contribution is flat and corresponds to a 0.24% contribution with a Gaussian-shaped distribution.

No pipetting contribution is present as the measurement is performed on the same physical sample. It is extremely important to compare the reproducibility of this instrument with the required reproducibility on ISO-8655-2 pipetting random errors reported in Table 1.

It can be reported that all volumes above 100 μL require testing random pipetting errors to a reproducibility of 0.3%, so if the measurement instrument has an intrinsic reproducibility of 0.24% the measured distribution for a pipette that performs according to the norms can be as large as 0.38%. After internal investigations, it is clear that there is no commercially available photometer performing in these experimental regimes with reproducibility that’s negligible with respect to the norms limit.

Therefore, it can be concluded that to assess the pipettes’ norm compatibility by photometry, the intrinsic measurement errors in the analysis should be included with the photometric method. Doing this offers the practical consequence of a limited sensitivity to assess the pipette compliance for pipettes that are performing just within the upper norms limit.

Well to Well Variations (Systematic Reproducability)

It is understandable to expect that the same sample could provide different absorbance measurements in relation to its well position in the microplate.

There are multiple potential reasons for a sample producing different absorbance measurement results. The first could be the different optical configuration of a well at the edges of a microplate when compared with the center.

The second could be irregularities or defects of the clear bottom and of the well plate walls. The third could be the internal design of the optical reader itself, in particular in relation to the number of light sources and detection heads used in a microplate scan.

Ultimately, such a test also includes the contribution from the pipetting itself, since pipetting samples will be affected by the unknown pipetting random error at the specific volume.

Considering these issues, experiments have been carried out with the Andrew robot by loading a constant amount of 150 μL of dye in the different locations and verified the uniformity of the reading. To minimize the random error generated by the optical reader the wells have been measured five consecutive times and the average result has been used in the analyses below.

The data shows a global variation of 0.51% among the microplate wells. There is no particular row or column dependence of the data, or evidence that the edges of the microplate, would lead to the production of different results.

Therefore, it can be concluded that the combination of pipetting random errors and the global photometer systematic effects account for 0.51%. Considering that the nominal random error for the pipetting itself at this volume is of the order of 0.2%, the analysis suggests that the main errors originate from the absorbance reader itself.

However, the performance of the microplate-based absorbance reader described above accurately represents the products that are commercially available and are practical for this purpose. So, it is confirmed that the previous conclusion indicating that the present performances of microplate-based absorbance readers constitute the main limitation for the random error assessment in the range of 100…200 µL where the pipette performances have random errors below their accuracy (see Table 1).

Buffer Light Absorption and Verification of the Method

Through experiments it has been possible to verify the hypothesis of the buffer-volume independence and the statement that the measurements only depend on the absolute amount of dye loaded into a well. This statement had some theoretical assumptions that needed experimental verification.

Andrew Alliance has experimentally verified the hypothesis of the buffer-volume independence and the statement that the measurements depend only on the absolute amount of dye loaded into a well.

This statement had some theoretical assumptions that necessitate experimental verification: particularly, the hypothesis of a negligible light absorption from the buffer, that the conical shape of the wells can be approximated as a cylinder (it is known that, for manufacturing-related reasons, wells sides have a draft angle of 1-2% to improve demolding), and that no unidentified effects (like a modulation of the refraction index by the dye concentration modifying the light reflections at the surfaces) affect the method.

Samples have been placed on the same microplate row to minimize the systematic reader error in the comparison, and 100 μL of dye have been loaded by an additional amount of water comprised between 0 and 200 μL, to achieve a total sample volume of between 100 μL and 300 μL.

The measurement is detailed in the figure above, along with the random error of 0.25%, which was predicted when considering previous measurements. It can be said with confidence that buffer volume inaccuracies introduced by pipetting have no effect on the measurements.

However, a dependence on the buffer volume is observed that becomes experimentally significant for differences of hundreds of microliters. For example, the same amount of dye diluted in a total volume of 300 μL produces a signal 0.45% lower than the same dye contained in a volume of 200 μL.

It is believed that the conical shape of the microplate wells has introduced this difference, and this was corrected accordingly for the systematic difference whenever required, for example when the nominal total volume of samples does not correspond. Indirectly, this measurement also proves the validity of the mixing procedure, as inaccuracies in mixing could result in well-to-well differences that aren’t observed.

Through the experiments, it was verified that not only is the average absorbance identical with varying amounts of buffer, but the random error of the measurements remain unaffected by the amount of buffer. This is an important validation to have made, as any inaccuracies induced by mixing will immediately impact the quality of data, and possibly create time-dependent effects.

Photometer Linearity

In order to use the calibration factor previously detailed, it is necessary to assess that the photometer behaves linearly as any calibrated weighting scale should. This verification was performed by adding a constant amount of sample into wells containing buffer, considering the fact that any systematic pipetting error would cancel, being the linearity essentially based on the digital counting of the number of subsequent additions.

Limiting to the data within the dynamic range of the photometer 0.3...3.0 OD, it is known from a previous experiment detailed in this report that the expected accuracy of the experimental configuration of microplate, pipetting and readout random errors, and the distribution of a constant volume of sample corresponds to 0.52% deviations. The data and a linear fit are shown in the following plot of absorbance vs. the number of fixed volumes of sample that has been added to the well.

It is important to note that data is background subtracted and that it has been previously demonstrated that the measurements are independent on the total volume of the liquid contained in the readout well. It is interesting to analyze the relative deviations of the individual measurements with respect to the straight-line fit.

The relative deviations are demonstrated in the plot below, highlighting the systematic and random error assessed for samples with uniform light absorption for each individual measurement.

The present plot shows that some systematic errors compete on random errors, being the points 2 to10 in row A and the points 11 to18 in row B. This confirms the evidence gathered during the analysis of signal differences from wells located in different positions on the microplate, larger than the intrinsic same-well reproducibility of the reader.

Considering the relative standard deviation of singular measurements from the straight line fit corresponds to 0.54% with a predictable, standard deviation of 0.51%, it can be concluded that the photometer is linear within the identified dynamic range and within the previously identified systematic errors already present in the analysis of samples at constant light absorbance.

This linearity measurement will affect the systematic error assessment, and in particular the comparison with the gravimetric method, and it is important to mention that the use of multiple dilution factors for the absorbance dye can trigger a situation where identical volumes are measured in different absorbance regimes, making the non-linearity discontinuities more visible.

This is certainly the case in the present experimental configuration for the measurements P2-2 and P20-2, P20-20 and P100-20, P100-100 and P200-100, in which the same liquid volume is measured by optical densities that differ by more than one order of magnitude. In these conditions, differences below 0.54% can originate from instruments themselves.

Gravimetric Method Errors

It has already been demonstrated that the photometer errors affect the resolution on the pipette performances. However, the choice of a classical semi-microbalance also impacts the performance of the measurements. The balance has a nominal readability of 0.01 mg and a nominal reproducibility between 0.04 mg and 0.05mg, dependent on the weighting conditions, and this reproducibility limitation introduces some constraints in the weighting of the small volume samples.

In particular, Table 1 indicates that below 20 μL volumes, the balance reproducibility limits the possibility of verifying the compliance to the ISO-8655 norm, since the balance measurement error has the same magnitude of the maximum random error of the data.

Through experiments, the reproducibility of the balance has been measured, on different days and in a normal laboratory environment, and concluded that the balance reproducibility corresponds to 0.05 mg. This factor is considered to be the dominant cause of measurement error for gravimetric measurements, since the Andrew robot operations run at a very constant speed, that enables them to statistically compensate for the water evaporation losses during the measurements (subtracted statistically). Additionally, a balance can be affected by a non-linear scale that could compromise the systematic error measurement.

It can be highlighted that these effects are modest in the small dynamic range of the measurements, and are generally compensated by the equal weight of the samples (dominated by the weight of the micro-tube and the support).

Common Sources of Errors

Choice of Tips

In the real world, the majority of users benefit from a large commercial offer of tips compatible with their pipettes to facilitate their individual needs in a cost effective manner. It is considered whether the liquid handling performances equivalent and independent from the tips are used.

There is debate among manufacturers over the tightness and mechanical compliance with the pipette end, as well as speculation over the fact that the different pipette tip shapes affect the hydrostatic forces acting on the liquid (11).

All these parameters are difficult to quantify through experiments reproducibly when a human user manipulates a pipette. However, the Andrew robot operates reproducibly and in constant conditions, always independently from the pipette tip involved.

Thus, a comparative experiment was performed and is described below, during which calibration experiments have been repeated, in sequence, by Andrew without changing the pipettes, firstly using Gilson Diamond tips (suggested tips for the pipettes used, see Table 2) and Eppendorf epTips. As the data shows, there is no observable effect within the measurement errors. Other tests and data for different brands are available on request.

Performances with Different Liquid Types

Biological samples rarely have the same physical properties as distilled water. This probability is decreased further when considering industrial and food-related samples. The addition of reagents and chemicals significantly alters the surface tension, pH, and viscosity, along with a wide range of other variables. This doesn’t even include suspensions, emulsions, or heterogeneous samples.

The dye-containing water samples have been laoded with 0.01% of TweenTM 20 (a common registered trade name for polysorbate 20). This is a minute amount of a well-known detergent typically used for improving the stability of a protein or 10% glycerol (v/v), which is often used to aid the storage of enzymes at low temperatures. The same calibration protocols described above have been repeated with these modified liquids. The results are as below.

The TweenTM 20 in the solution generates bubbles that affect both the systematic and the random errors. The TweenTM 20 reduces the surface tension of solutions but also creates microscopic bubbles that can become trapped inside the tip and cannot be ejected when the second stop is pressed.

When a tip is used multiple times, the trapped bubbles can also remain in place for an extended duration of time and ejected randomly at any dispensing action. These results show that in the presence of TweenTM 20, the dispensed volume is systematically larger when compared to the nominal volume set. This significant difference in volume is most probably created by the different contact angle of the modified solution against the tip’s polypropylene material.

The results gathered from experiments with the 10% glycerol differ significantly. This is because glycerol only affects the systematic errors, leaving random errors untouched.

The volumes dispensed are slightly bigger that the nominal volumes but in a very reproducible manner. However, the density of the glycerol-based solution is 1.021 g/cm3 at the working temperature (approaching 25 °C), while distilled water has a density of 0.997 g/cm3: a difference of 2.4%, which certainly affects the systematic performances of pipettes as the additional weight of the liquid changes the internal air pressure due to a higher Bernoulli force.

It can thus be concluded, including water-based samples, the physical properties of a liquid can have substantial and practically unpredictable consequences on the volumetric performances of a pipette, easily exceeding the limits of the ISO-8655 norms.

Experiment Repeatability

A large series of experiments has been conducted, both consecutively over the course of the same day and separately on different days. It has been possible to determine one clear advantage of photometry, which is related to the substantially reduced dependence from environmental factors, validated by the good reproducibility of the data obtained for the same setup.

The dispersion of three consecutive experiments by the same Andrew system and the same pipette set is shown in Table 7 below for the full volume range. This data is representative of the repeatability of one Andrew system in the same laboratory. It can confidently be concluded that, within the measurement errors, the system systematically performs within 1% tolerance with respect to the nominal set volume, with an exception in the region below 2 μL. In this region, pipette performances degrade to approximately 5% (but corresponding to the minute amount of a <40 nL reproducibility).

Summary of Measurement Errors

Essentially, photometric and gravimetric methods are good techniques for the assessment of pipetting performances with different limitations and advantages. Photometry has a limited best reproducibility of approximately 0.3%. This makes the assessment of pipette performances challenging in the region of 100-200 μL, where pipettes have a reproducibility of the same order of magnitude.

However, its limited sensitivity to evaporation makes it a notably practical option for the small volumes region, especially when considering the overall convenience of the reader.

On the contrary, a semi-microbalance excels in the large volume region, being limited by an absolute weighting error of 0.05 mg corresponding to 0.05 μL, totally negligible above 100 μL. However, this method is unfortunately not practical in normal lab conditions when it comes to measurements below a few microliters. Overall, a weight measurement is the only method that can provide an absolute calibration scale, which is an aspect that justifies its widespread use in calibration laboratories and maintenance services.

6. Other Sources of Errors

As pipettes are primarily used by humans, it is not possible to avoid assessing the errors induced by a human operator.

Skilled Pipetting Operator Vs. Andrew

A number of manual experiments have been performed reproducing the operations performed by the Andrew robot. Through this it has been possible to measure the pipetting performances by photometric methods and by weighting.

The operator, who is highly trained in conventional laboratory procedures and GLPs, applied every precision-enhancing technique suggested by the pipette manufacturers for calibration purposes, including tip pre-wetting. The results by photometry are reported below:

Graphically comparing the data produced by a skilled human operator and Andrew itself by representing all data in the same plot (Fig. 9) shows it is clear that in the majority of cases, random errors by Andrew (green dots) are below the random errors from the operator (blue dots).

Consistently, systematic errors are correlated (yellow and red dots). This definitely proves that Andrew Alliance has been successful in their endeavors to tune Andrew to behave as a skilled human operator. As such, Andrew Alliance conclude that Andrew can behave like a skilled operator would do, exploiting at the limit the ideal pipetting performances with high reproducibility.

Daily Operations in the Lab: A Different Story

Although not strictly connected with intrinsic pipetting performances, it is essential to reinforce the fact that there is a plethora of aspects that affect pipetting performance in the laboratory, with most of them being related to daily practices. These include:

  • Not performing pre-wetting account for errors up to 2% (12);
  • Uneven rhythm and timing during pipetting can increase errors by 1.5%;
  • Non-consistent and incorrect tip immersion depth into the liquid can account for up to 1% variations;
  • A deviation from verticality of the piston axis of 20% can induce a 1% error (13)
  • Approximate sensing of the piston “first stop” can introduce systematic errors of 0.5%;
  • Parallax errors and incorrect volume setting can account for errors up to 0.5%. It is enough to keep the pipette in front of the eyes’ field of vision, and to always set the volume by exceeding the desired target by three units and moving back the counter to the nominal volume to avoid hysteresis;
  • Low quality tips can induce leaks accounting for 0.5 to 50% losses.


Numerous experiments have been defined, performed and analyzed to assess the performance of pipettes, using the Andrew robot to perform pipetting in a precisely reproducible manner, thereby removing the uncertainty induced by a human operator.

The gravimetric and photometric methods have been compared and concluded that in normal lab conditions and utilizing commercial equipment, the two methods are characterized by different limitations and consequent performances.

The photometric method is efficiently automated and, when using Andrew, performs pipette calibration with no need for oversight from the user. The photometric method excels in testing low volume pipettes effectively when compared to the ISO standard, but it is marginal in the compliance assessment of large volume pipettes. This is due to the limited reproducibility of commercial photometers.

In comparison, the gravimetric method provides the best data for large volume pipettes and provides an absolute calibration scale. However, it presents significant challenges when used in the low-volume range. Overall, the gravimetric method is less suited to automation and in the low-volume range, expensive and delicate equipment is needed. Furthermore, its sensitivity to evaporation does not make it a practical option in the humidity and temperature conditions of a typical laboratory.

After a considered review of errors and performances in the conditions of a normal lab, we conclude that photometry is a valuable and importantly practical method to validate volumetric performances in the range 0.5 μL-1000 μL, and that this procedure can be fully automated by means of the Andrew robot when complemented by a compact, cost-effective photometer. Independently of the preferred assessment method, it is concluded that the Andrew robot performs, in standard laboratory conditions, as the best pipetting operator would do.

Table 1. Minimal Pipetting Performances as Indicated by the Norm ISO-8655-2 (2002)

Nominal volume Maximum permissible systematic error Maximum permissible random error
[μl] [%] [μl] [%] [μl]
1 ± 5 ± 0.05 ± 5 ± 0.05
2 ± 4 ± 0.08 ± 2 ± 0.04
5 ± 2,5 ± 0.125 ± 1.5 ± 0.075
10 ± 1.2 ± 0.12 ± 0.8 ± 0.08
20 ± 1 ± 0.2 ± 0.5 ± 0.1
50 ± 1 ± 0.5 ± 0.4 ± 0.2
100 ± 0.8 ± 0.8 ± 0.3 ± 0.3
200 ± 0.8 ± 1.6 ± 0.3 ± 0.6
500 ± 0.8 ± 4.0 ± 0.3 ± 1.5
1000 ± 0.8 ± 8.0 ± 0.3 ± 3.0
2000 ± 0.8 ± 16.0 ± 0.3 ± 6.0
5000 ± 0.8 ± 40.0 ± 0.3 ± 15.0
10000 ± 0.6 ± 60.0 ± 0.3 ± 30.0


Blank table

Volume (μL) Pipette
300 P1000
200 P1000
200 P200
150 P200
100 P200
100 P100
50 P100
20 P100
20 P100
10 P20
5 P20
2 P20
2 P2
1 P2
0.5 P2


Table 2. Measured photometric performances for a set of 5 pipettes measured by Andrew.

  Expected Volume
Average Volume
Relative Inaccuracy
Systematic Error
Random Error
P2-0.5 0.5 0.50 -0.5% 3.3% 0.00 0.02
P2-1 1 1.01 0.8% 0.9% 0.01 0.01
P2-2 2 1.95 -2.6% 0.7% -0.05 0.01
P20-2 2 2.10 4.9% 0.7% 0.10 0.02
P20-5 5 5.12 2.4% 0.2% 0.12 0.01
P20-10 10 10.15 1.5% 0.7% 0.15 0.07
P20-20 20 20.04 0.2% 0.1% 0.04 0.03
P100-20 20 20.75 3.8% 0.5% 0.75 0.11
P100-50 50 50.48 1.0% 0.2% 0.48 0.13
P100-100 100 99.55 -0.5% 0.2% -0.45 0.22
P200-100 100 100.86 0.9% 0.4% 0.86 0.46
P200-150 150 149.79 -0.1% 0.2% -0.21 0.33
P200-200 200 199.34 -0.3% 0.3% -0.66 0.51
P1000-200 200 201.62 0.8% 0.9% 1.62 1.83
P1000-300 300 301.04 0.3% 0.4% 1.04 1.19


Table 3. Measured gravimetric performances for a set of 5 pipettes measured by Andrew.

  Expected Volume
Average Volume
Relative Inaccuracy
Systematic Error
Random Error
P2-0.5 0.50 0.42 -16.0% 0.09 -0.08 0.04
P2-1 1.00 0.96 -4.0% 0.03 -0.04 0.03
P2-2 2.00 1.93 -3.5% 0.03 -0.07 0.07
P20-2 2.00 1.98 -1.0% 0.05 -0.02 0.10
P20-10 10.00 9.80 -2.0% 0.01 -0.20 0.09
P20-20 20.00 19.83 -0.9% 0.01 -0.17 0.15
P100-20 20.00 20.24 1.2% 0.01 0.24 0.17
P100-50 50.00 50.08 0.2% 0.00 0.08 0.06
P100-100 100.00 100.09 0.1% 0.00 0.09 0.12
P200-100 100.00 100.19 0.2% 0.00 0.19 0.10
P200-150 200.00 200.41 0.2% 0.00 0.41 0.12
P200-200 200.00 199.57 -0.2% 0.00 -0.43 0.24
P1000-500 500.00 501.05 0.2% 0.00 1.05 0.57
P1000-1000 1000.00 1000.63 0.1% 0.00 0.63 0.38


Table 5. 2D map of a microplate (% variations with respect to average signal)

  A B C D E F G H I J K L
1 -0.8% -0.5% -0.6% -0.7% -0.5% -0.6% -0.9% -0.2% -0.3% 0.3% 0.1% 0.6%
2 -0.1% -0.4% -0.2% -0.4% -0.1% -0.6% -0.5% -0.2% 0.6% -0.2% 0.0% -1.1%
3 -0.1% 0.1% 0.2% 0.0% 0.0% -0.2% -0.4% -0.4% -0.2% -0.9% 0.5% 0.5%
4 0.2% -0.3% 0.2% -0.1% -0.2% -0.1% -0.2% -0.2% 0.2% 0.1% 0.3% 0.0%
5 0.1% -0.1% 0.2% -0.2% -0.1% 0.3% -0.4% 0.1% 0.5% 0.0% 0.1% 0.4%
6 -0.3% 0.0% 0.1% 0.2% -0.1% 0.2% 0.3% 0.6% 0.5% 0.2% 0.4% 0.1%
7 0.2% 0.0% -1.1% 0.4% 0.2% 0.2% 0.4% 0.4% 0.5% 0.7% 0.5% 0.3%
8 0.1% 0.2% 0.1% 0.0% 0.0% 0.2% 0.0% 0.4% 0.7% -0.1% 0.6% 0.6%


Table 4. Calibration data obtained with eppendorf eptips tips.

  Expected Volume
Average Volume
Relative Inaccuracy
Systematic Error
Random Error
P2-0.5 0.5 0.47 -5.7% 2.1% -0.03 0.01
P2-1 1 0.97 -2.8% 0.8% -0.03 0.01
P2-2 2 1.94 -2.9% 0.4% -0.06 0.01
P20-2 2 2.09 4.4% 2.0% 0.09 0.05
P20-5 5 5.12 2.5% 0.6% 0.12 0.03
P20-10 10 10.12 1.2% 0.4% 0.12 0.05
P20-20 20 20.12 0.6% 0.3% 0.12 0.07
P100-20 20 20.71 3.5% 0.4% 0.71 0.09
P100-50 50 50.42 0.8% 0.3% 0.42 0.15
P100-100 100 99.70 -0.3% 0.3% -0.30 0.27
P200-100 100 101.09 1.1% 0.2% 1.09 0.22
P200-150 150 150.58 0.4% 0.4% 0.58 0.56
P200-200 200 199.82 -0.1% 0.2% -0.18 0.33
P1000-200 200 199.81 -0.1% 0.3% -0.19 0.53
P1000-300 300 297.62 -0.8% 0.5% -2.38 1.49


Table 5. Calibration data obtained with liquid containing 0.01% of Tween™ 20

  Expected Volume
Average Volume
Relative Inaccuracy
Systematic Error
Random Error
P2-0.5 0.5 0.68 35.2% 11.6% 0.18 0.08
P2-1 1 1.23 23.2% 5.8% 0.23 0.07
P2-2 2 2.17 8.4% 3.7% 0.17 0.08
P20-2 2 2.47 23.7% 6.5% 0.47 0.19
P20-5 5 5.82 16.4% 1.0% 0.82 0.06
P20-10 10 10.98 9.8% 0.6% 0.98 0.07
P20-20 20 21.08 5.4% 0.6% 1.08 0.12
P100-20 20 22.53 12.7% 0.5% 2.53 0.13
P100-50 50 52.35 4.7% 0.4% 2.35 0.19
P100-100 100 102.23 2.2% 0.4% 2.23 0.28
P200-100 100 103.32 3.3% 1.3% 3.32 1.35
P200-150 150 153.87 2.6% 0.4% 3.87 0.57
P200-200 200 201.17 0.6% 1.8% 1.17 3.65
P1000-200 200 206.09 3.0% 0.5% 6.09 1.04
P1000-300 300 314.95 5.0% 1.7% 14.95 5.51


Table 6. : Calibration data obtained with liquid containing 10% glycerol

  Expected Volume
Average Volume
Relative Inaccuracy
Systematic Error
Random Error
P2-0.5 0.5 0.55 9.1% 4.2% 0.05 0.02
P2-1 1 1.05 5.5% 0.7% 0.05 0.01
P2-2 2 2.02 0.9% 0.9% 0.02 0.02
P20-2 2 2.16 8.0% 0.8% 0.16 0.02
P20-5 5 5.24 4.9% 0.9% 0.24 0.05
P20-10 10 10.34 3.4% 0.6% 0.34 0.06
P20-20 20 20.52 2.6% 0.3% 0.52 0.07
P100-20 20 21.32 6.6% 0.4% 1.32 0.09
P100-50 50 51.68 3.4% 0.3% 1.68 0.16
P100-100 100 102.18 2.2% 0.3% 2.18 0.29
P200-100 100 103.82 3.8% 0.5% 3.82 0.51
P200-150 150 154.34 2.9% 0.1% 4.34 0.19
P200-200 200 205.45 2.7% 0.2% 5.45 0.45
P1000-200 200 203.62 1.8% 1.4% 3.62 2.98
P1000-300 300 301.38 0.5% 0.7% 1.38 2.27


Table 7. Experiment repeatability (same Andrew, same Pipettes)

  Experiment repeatability (%) Experiment repeatability (μl)
P2-0.5 6.80% 0.03
P2-1 2.08% 0.02
P2-2 2.07% 0.04
P20-2 0.72% 0.02
P20-5 0.70% 0.04
P20-10 0.53% 0.05
P20-20 0.38% 0.08
P100-20 0.45% 0.09
P100-50 0.41% 0.20
P100-100 0.77% 0.77
P200-100 0.62% 0.63
P200-150 0.60% 0.90
P200-200 0.22% 0.44
P1000-200 0.83% 1.66
P1000-300 0.42% 1.24


Table 8. Photometric data from a skilled operator (same Protocol)

test Expected Volume
Average Volume
Relative Inaccuracy
Systematic Error
Random Error
P2-0.5 0.5 0.52 3.2% 3.4% 0.02 0.02
P2-1 1 1.05 4.9% 1.5% 0.05 0.02
P2-2 2 1.99 -0.7% 0.6% -0.01 0.01
P20-2 2 2.12 5.9% 0.8% 0.12 0.02
P20-5 5 5.16 3.1% 0.6% 0.16 0.03
P20-10 10 10.19 1.9% 1.1% 0.19 0.11
P20-20 20 20.20 1.0% 0.3% 0.20 0.05
P100-20 20 20.63 3.2% 1.4% 0.63 0.30
P100-50 50 50.18 0.4% 0.3% 0.18 0.16
P100-100 100 99.36 -0.6% 0.3% -0.64 0.26
P200-100 100 100.87 0.9% 0.4% 0.87 0.37
P200-150 150 149.43 -0.4% 0.2% -0.57 0.37
P200-200 200 199.39 -0.3% 0.3% -0.61 0.56
P1000-200 200 202.66 1.3% 0.3% 2.66 0.53
P1000-300 300 297.57 -0.8% 0.3% -2.43 0.82


Balance scale from the Egyptian Book of the Dead.

Figure 1: Balance scale from the Egyptian Book of the Dead.

Andrew, the pipetting robot using manual pipettes.

Figure 2: Andrew, the pipetting robot using manual pipettes.

Normalized representation of pipetting performances (Andrew).

Figure 3: Normalized representation of pipetting performances (Andrew).

Relative error as a function of absorbance (Same Well).

Figure 4: Relative error as a function of absorbance (Same Well).

Signal variations for wells additionally loaded with different amounts of buffer.

Figure 6: Signal variations for wells additionally loaded with different amounts of buffer.

Linear fit to data generated by discrete additions of dye.

Figure 7: Linear fit to data generated by discrete additions of dye.

Deviation from linearity, in view of expected errors.

Figure 8: Deviation from linearity, in view of expected errors.

Normalized representation of pipetting performances (Operator vs. Andrew)

Figure 9: Normalized representation of pipetting performances (Operator vs. Andrew)

About Andrew Alliance S.A.

Andrew Alliance is an independent, privately financed company, based in Geneva, Boston and Paris. The company was created in March 2011.

Andrew Alliance is dedicated to advance science by working with scientists to create a new class of easy-to-use robots and connected devices that take repeatability, performance, and efficiency of laboratory experiments to the level required by 21st-century biology.

Start with meeting customer needs, end with customer feedback.

Andrew Alliance delivers solutions that are focused on customer needs, both today and in the future. Our products are manufactured to the highest standards, using a range of carefully selected, proven, and sustainable technologies, that ensure both high performance and reliability. We actively seek continuous customer feedback, in order to guarantee the best possible design outcomes.

Sponsored Content Policy: publishes articles and related content that may be derived from sources where we have existing commercial relationships, provided such content adds value to the core editorial ethos of News-Medical.Net which is to educate and inform site visitors interested in medical research, science, medical devices and treatments.

Last updated: Jul 22, 2019 at 7:57 AM


Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    Andrew Alliance S.A.. (2019, July 22). Developing & Automating Pipetting with Pipetting Robots. News-Medical. Retrieved on September 23, 2019 from

  • MLA

    Andrew Alliance S.A.. "Developing & Automating Pipetting with Pipetting Robots". News-Medical. 23 September 2019. <>.

  • Chicago

    Andrew Alliance S.A.. "Developing & Automating Pipetting with Pipetting Robots". News-Medical. (accessed September 23, 2019).

  • Harvard

    Andrew Alliance S.A.. 2019. Developing & Automating Pipetting with Pipetting Robots. News-Medical, viewed 23 September 2019,

Other White Papers by this Supplier